Writing for the Journal of Orthopaedic Research


  • 1 Department of Biomechanics and Biomaterials, Hospital for Special Surgery, New York, New York 10021-4892, USA. [email protected]
  • PMID: 10459750
  • DOI: 10.1002/jor.1100170402
  • Orthopedics*

U.S. flag

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

  • Publications
  • Account settings
  • Advanced Search
  • Journal List
  • J Orthop Traumatol
  • v.9(1); 2008 Mar

Logo of jortr

How to write an original article for the Journal of Orthopaedics and Traumatology

L. pierannunzii.

Third Division of Orthopaedics and Trauma, Istituto Ortopedico G. Pini, Via Pini 9, Milano, Italy

The abstract is the precise summary of the article, not a preface. As Baue wrote in a popular editorial of the Archives of Surgery in 1979, “writing a good abstract is not abstract writing” [ 6 ]. The main data have to be represented, as they allow readers to understand contents clearly. Sentences like “The paper reports...” or “The authors describe...” have to be avoided as well as any generic statements.

In order to help writers avoid generalities, the recently revised version of this journal’s “instructions to authors” [ 4 ] requires the abstract to be no longer than 300 words and structured in 4 paragraphs: the Background declares the hypothesis, the Materials and methods impart the study design and quote the relevant numerical features of the samples, the Results report the main data and their statistical significance, the Conclusions state whether the hypothesis is verified or not. One or two sentences per paragraph are usually sufficient.

The abstract is frequently recommended to be written after the text, as “the process of writing changes thought and perhaps even purpose” [ 7 ]. Nonetheless, in my view, preparing the abstract first is a useful exercise that forces the authors to organize their thoughts and guides the organization of the article. However, the abstract should be always revised after the manuscript has been completed.

An original article (OA) is the publication of a study. Thus, it might be alternatively named “study report”, by analogy with “case report”. The Journal of Orthopaedics and Traumatology welcomes OAs based on clinical or preclinical studies relevant to the musculoskeletal system.

Although medical research methodology is not the object of the following writing, proper planning and performing of investigations are necessary requirements for any good article. Three points have to be clear in the author’s mind: the hypothesis, the materials and methods through which the hypothesis is tested, and the results [ 1 ]. Only studies logically structured according to this sequence will be successfully transformed into worthwhile reports.

Precisely discussing the level of evidence is beyond the aims of the present editorial, but authors need to be aware that the design of their study affects the quality of the scientific information that is conveyed. Clinical studies may be roughly listed in order of decreasing evidence, as follows: randomized controlled trials (RCT), non-RCT prospective comparative studies, retrospective comparative studies, case-control studies and, lastly, case series [ 2 ].

Case series, characterized by a low level of evidence, deserve publication only if the sample size and follow-up are strictly adequate. Sample size has to be at least similar or possibly larger than previously reported series. Minimum follow-up (not the mean follow-up) has to be long enough to allow observations of steady clinical and radiological outcomes. This means, for instance, 1 year for long bones fractures and 2 years for articular fractures with possible secondary degenerative joint disease.

Special features are requested if the purpose of a case series is to estimate long-term survival of patients in relation to death or recurrency (e.g. musculoskeletal tumours), of organs in relation to prosthetic replacement (e.g. joints after corrective surgery or articular fracture reconstruction), and of prostheses in relation to failure (e.g. total hip or knee replacement). These series need not only an adequate minimum follow-up (generally several years), but also a proper survival analysis.

As far as prosthetic case series (standard total hip or knee arthroplasty) are concerned, the Editorial Committee sets a minimum follow-up of 8 years and a minimum sample size of 150 implants. Survival analysis should be preferably performed with the Kaplan-Meier estimator [ 3 ]. New and original prosthetic designs may represent an exception, for which smaller series and shorter follow-up could be accepted if the literature does not contain larger or longer studies.

OAs submitted to the Journal of Orthopaedics and Traumatology have to be arranged in 5 sections: Abstract, Introduction, Materials and methods, Results and Discussion. A limit of 3500 words is far beyond the needs of most good papers [ 4 ].

The medical writing style, officially set by the Manual of Style of the American Medical Association, should be grammatically correct, clear and not redundant. In one word, essential. [ 5 ]

Any possible conflict of interest has to be disclosed. Otherwise, the authors should clearly declare that no funds were received in support of the research that researchers had no financial interests related to it.


The Introduction is a critical section, as it needs to be finely balanced to allow a proper approach to the subject matter without anticipating the contents of the subsequent parts. Important topics of the Introduction are the state of the art and the hypothesis (or hypotheses). The state of the art should be briefly described with essential papers. Deeply commenting on a single reference is better postponed to the Discussion, since the purpose of this section is to inform, not to discuss.

The hypothesis has to be clearly expressed in the last paragraph, and its relevance should be logically deducible from the previous state of the art description. In other words, the Introduction aims at showing that a problem exists, and that previous investigations did not offer any adequate solutions, so justifying the reported research [ 8 ].

Materials and methods

This section should contain the detailed description of the study. The more careful this description is, the more reliable the results are.

As for clinical studies, the following contents cannot be omitted:

  • Declaration that the study was carried out in accordance with the World Medical Association’s Declaration of Helsinki [ 8 ], that it was approved by the institutional ethics committee and that all the patients gave informed consent to be enrolled.
  • Study design. Is it a prospective or retrospective study? Is it a case-control study or a case series? Is it double-blinded, single-blinded or open?” Were the patients samples randomized or not? These are some of the questions that need to be answered here.
  • Inclusion and exclusion criteria.
  • Characteristics of the patients (mean age and range, male-to-female ratio, diagnoses, confounding variables, etc).
  • Outcome measures (clinical variables, radiological variables, combined scores, etc) should be broadly accepted in the representative literature. In case of a different choice, this decision has to be justified.
  • Statistical methods.
  • Significance level (e.g. α = 0.05) and power (e.g. β = 0.8).

As previously indicated, cases series regarding total joint replacement and malignant tumours need a survival analysis.

The purpose of this section is to provide numerical data without comments [ 10 ].

Clinical studies should always report the number of cases lost to follow-up. Variables have to be reported as averages and 95% confidence intervals. Units should be always indicated and abbreviated according to the metric or SI system. Frequencies (e.g. frequency of complications) have to be described both by absolute number and percentage, the latter being in parentheses, e.g. 6 (2.5%).

If the sample size is small (up to 20 cases), a detailed table is requested that displays the most remarkable measured variables of each case. Authors should remember that such a table ought to be available anyway, as referees might ask for it during the reviewing process.

Since the aim of tables is to save space, they should be avoided if the same data may be presented more concisely in the text or if they just duplicate text contents.

Results that are relevant to the hypothesis have to be associated with their statistical significance. The exact p value has to be reported close to “significant” or “non-significant”, while “ p < α” cannot be accepted, as it does not allow readers to understand the true risk of type I error (the risk of observing a difference that does not exist) and reviewers to check the calculations.

Results not relevant to the hypothesis have to be reported only if they show unexpected findings or might be useful for further investigations. Otherwise they will divert attention from the main results.

The Discussion is the section in which previously reported results are discussed, not repeated nor summarized. Here, the author is asked to achieve four fundamental goals:

  • Compare the study to the relevant literature,
  • Acknowledge possible weak points,
  • Draw conclusions about the hypothesis (verified or not), and
  • State the clinical relevance of these results.

The first task is obtained through a careful review of available studies regarding the subject matter, which should be briefly referred to in the text, without getting lost in excessively detailed analyses. Discrepancies and unexpected findings have to be explained, or at least attempted to be explained.

The second issue is an essential step of any scientific paper. Author who do not highlight limitations of their study (bias, short follow-up, small sample size, etc.) show superficiality and lack of self-criticism, so compromising their own credibility and reliability of the results.

The third point is the manuscript’s conclusion, where the authors are expected to state if the experimental hypothesis was verified or not on the basis of the results. This cannot stand alone without the fourth part, in which the clinical relevance of the conclusions are set forth.

The multiple issues of the Discussion usually make it a long section. Thus, authors should pay attention to avoid repetitions, redundancy and digressions.

Every statement that is not proved by the results of the study nor can be logically drawn by a previous one needs to be supported by a specific reference. References should be pertinent and recent. No mentions to personal communications or to proceedings older than 3 years are permitted, in order to allow readers to consult the sources easily.

Up to 50 references are permitted in OAs of the Journal of Orthopaedics and Traumatology . The authors should remember to check the intructions to authors for their structure and order of citation [ 4 ].

Writing an OA for the Journal of Orthopaedics and Traumatology , as for any other peer-reviewed journal, first requires a reliable study. Poorly posed hypotheses, badly planned study protocols, inaccurate data collection and wrong statistical analyses compromise the quality of the final manuscript much more than do writing errors, often irreversibly.

On the other hand, if a sound study was performed, transforming it into a good article is barely a matter of form. Although the suggestions provided here are intended to help authors with manuscripts preparation, an effective writing style is mainly an achievement of experience. The only way to shorten the learning curve is by reading. Every author and especially the future ones should keep in mind that the best medical writers are the most assiduous medical readers.

Europe PMC requires Javascript to function effectively.

Either your web browser doesn't support Javascript or it is currently turned off. In the latter case, please turn on Javascript support in your web browser and reload this page.

Search life-sciences literature ( 43,401,315 articles, preprints and more)

  • Available from publisher site using DOI. A subscription may be required. Full text
  • Citations & impact
  • Similar Articles

Writing for the Journal of Orthopaedic Research.

Author information, affiliations.

  • All authors 1. Department of Biomechanics and Biomaterials, Hospital for Special Surgery, New York, New York 10021-4892, USA. [email protected]

ORCIDs linked to this article

  • Buckwalter JA | 0000-0003-4308-7583

Journal of Orthopaedic Research : Official Publication of the Orthopaedic Research Society , 01 Jul 1999 , 17(4): 459-466 https://doi.org/10.1002/jor.1100170402   PMID: 10459750 


Full text links .

Read article at publisher's site: https://doi.org/10.1002/jor.1100170402


Articles referenced by this article (5)

Scientific credibility requires complete presentation of methods.

Buckwalter JA , Wright TM , Frank CB , Martin RB , Sandell LJ , Trippel SB

J Orthop Res, (2):161 1997

MED: 9167616

: The Elements of Style, 3rd ed, p 23. New York, Macmillan Publishing, 1979

Style manual committee, council of biology editors [eds]: scientific style and format: the cbe manual for authors, editors, and publishers, 6th ed. cambridge, cambridge university press, 1994, title not supplied.

J Orthop Res 1997

: Essentials of Writing Biomedical Research Papers. New York, McGraw-Hill, 1991

Citations & impact , impact metrics, alternative metrics.

Altmetric item for https://www.altmetric.com/details/25203035

Article citations

Why, when, who, what, how, and where for trainees writing literature review articles..

Koons GL , Schenke-Layland K , Mikos AG

Ann Biomed Eng , 47(11):2334-2340, 21 May 2019

Cited by: 1 article | PMID: 31115720 | PMCID: PMC6842086

Free full text in Europe PMC

Similar Articles 

To arrive at the top five similar articles we use a word-weighted algorithm to compare words from the Title and Abstract of each citation.

Writing for Clinical Orthopaedics and Related Research.

Clin Orthop Relat Res , (413):1-7, 01 Aug 2003

Cited by: 7 articles | PMID: 12897589

The challenge of success for the Journal of Orthopaedic Research.

Buckwalter JA

J Orthop Res , 13(3):307-308, 01 May 1995

Cited by: 0 articles | PMID: 7602390

Plagiarism: an unprofessional and dishonest trend creeping into modern writing of research and review papers.

Stephens MM

Foot Ankle Surg , 15(2):57, 24 Jan 2009

Cited by: 0 articles | PMID: 19410168

Writing for publication.

Axone , 22(4):36-39, 01 Jun 2001

Cited by: 3 articles | PMID: 14621504

Disseminating nursing knowledge--a guide to writing for publication.

Int J Psychiatr Nurs Res , 10(3):1147-1155, 01 May 2005

Cited by: 7 articles | PMID: 15960243

Europe PMC is part of the ELIXIR infrastructure

Contribute to the future of Europe PMC. Take part in our user survey!

VIVO Weill Cornell Medical College

This site uses HTML elements that are not recognized by Internet Explorer 8 and below in the absence of JavaScript. As a result, the site will not be rendered appropriately. To correct this, please either enable JavaScript, upgrade to Internet Explorer 9, or use another browser. Here are the instructions for enabling JavaScript in your web browser .

Journal of orthopaedic research : official publication of the Orthopaedic Research Society Journal

Publication venue for.

  • The 2023 Orthopaedic Research Society's International Consensus Meeting on musculoskeletal infection: Summary from the host immunity section. 2023
  • A multimodal assessment of cementless tibial baseplate fixation using radiography, radiostereometric analysis, and magnetic resonance imaging. 2023
  • Bone-ACL-bone allograft for anterior cruciate ligament reconstruction: Short-term evaluation in a rabbit model with microcomputed tomography. .  41. 2023
  • MAVRIC based T2 mapping assessment of infrapatellar fat pad scarring in patients with total knee arthroplasty. .  41. 2022
  • Adaptive immune responses in patients requiring revision after Total Knee Arthroplasty. 2022
  • Dermal ultrasound measurements for bone quality assessment : An investigation of advanced glycation endproducts derived from confocal fluorescence microscopy. .  41. 2022
  • Assessment of reporting practices and reproducibility potential of a cohort of published studies in computational knee biomechanics. 2022
  • Evaluation of sex differences in rodent anterior cruciate ligament injury. 2022
  • A novel mouse model of hindlimb joint contracture with 3D-printed casts. .  40. 2022
  • The role of Meteorin-like in skeletal development and bone fracture healing. .  40. 2022
  • High resolution ultrasound for the evaluation of radial nerve pathology adjacent to metallic hardware. 2022
  • Assessment of osteoarthritis functional outcomes and intra-articular injection volume in the rat anterior cruciate ligament transection model. .  40. 2022
  • Trabecular volumetric bone mineral density of the occipital bone at preferred screw placement sites measured by quantitative computed tomography. .  40. 2021
  • Neoplastic synovial lining cells that coexpress podoplanin and CD90 overproduce CSF-1, driving tenosynovial giant cell tumor. .  40. 2021
  • Targeted transcriptomic analyses of RNA isolated from formalin-fixed and paraffin-embedded human menisci. 2021
  • Effect of varus alignment on the bone-implant interaction of a cementless tibial baseplate during gait. 2021
  • Identifying alternative antibiotics that elute from calcium sulfate beads for treatment of orthopedic infections. 2021
  • Kinematic pelvic tilt during gait alters functional cup position in total hip arthroplasty. .  40. 2021
  • The cervical spine demonstrates less postoperative bone loss than the lumbar spine. 2021
  • Knee cartilage T 2 relaxation times 3 months after ACL reconstruction are associated with knee gait variables linked to knee osteoarthritis. .  40. 2021
  • Histologic and molecular features in pathologic human menisci from knees with and without osteoarthritis. 2021
  • Chronic subacromial impingement leads to supraspinatus muscle functional and morphological changes: Evaluation in a murine model. 2020
  • Utility of lavage in addition to native fluid collection during fluoroscopically guided joint aspiration in infection diagnosis. .  39. 2020
  • Biomechanical evaluation of total ankle arthroplasty. Part I: Joint loads during simulated level walking. .  39. 2020
  • Systemic osteoprotegerin does not improve peri-implant bone volume or osseointegration in rabbits. .  39. 2020
  • Biomechanical evaluation of total ankle arthroplasty. Part II: Influence of loading and fixation design on tibial bone-implant interaction. .  39. 2020
  • Rate of infection following revision anterior cruciate ligament reconstruction and associated patient- and surgeon-dependent risk factors: Retrospective results from MOON and MARS data collected from 2002 to 2011. .  39. 2020
  • Alendronate enhances osseointegration in a murine implant model. .  39. 2020
  • Multiparametric MRI characterization of knee articular cartilage and subchondral bone shape in collegiate basketball players. .  39. 2020
  • Obesity and load-induced posttraumatic osteoarthritis in the absence of fracture or surgical trauma. .  39. 2020
  • A geometric ratio to predict the flexion gap in total knee arthroplasty. .  38. 2020
  • Proximal-distal shift of the center of rotation in a total wrist arthroplasty is more than twice of the healthy wrist. .  38. 2020
  • Expression of alarmins in a murine rotator cuff tendinopathy model. .  38. 2020
  • Multivariate use of MRI biomarkers to classify histologically confirmed necrosis in symptomatic total hip arthroplasty. .  38. 2020
  • 1.5 T magnetic resonance imaging generates accurate 3D proximal femoral models: Surgical planning implications for femoroacetabular impingement. .  38. 2020
  • American Society for Bone and Mineral Research-Orthopaedic Research Society Joint Task Force Report on Cell-Based Therapies - Secondary Publication. .  38. 2020
  • The continuum of hip range of motion: From soft-tissue restriction to bony impingement. .  38. 2020
  • Amniotic Suspension Allograft Modulates Inflammation in a Rat Pain Model of Osteoarthritis. .  38. 2019
  • In vitro test methods for seating and fretting corrosion behavior of modular metal-on-metal acetabular tapers. .  38. 2019
  • The biomechanics of subscapularis repair in reverse shoulder arthroplasty: The effect of lateralization and insertion site. .  38. 2019
  • Combined Injury to the ACL and Lateral Meniscus Alters the Geometry of Articular Cartilage and Meniscus Soon After Initial Trauma. .  38. 2019
  • The MRL/MpJ Mouse Strain Is Not Protected From Muscle Atrophy and Weakness After Rotator Cuff Tear. .  38. 2019
  • Skin Ultrasound Measurement as a Potential Marker of Bone Quality: A Prospective Pilot Study of Patients undergoing Lumbar Spinal Fusion. .  37. 2019
  • Predicting Carpal Bone Kinematics Using an Expanded Digital Database of Wrist Carpal Bone Anatomy and Kinematics. .  37. 2019
  • Identification of Inflammatory Mediators in Tendinopathy Using a Murine Subacromial Impingement Model. .  37. 2019
  • Effects of a Medial Knee Unloading Implant on Tibiofemoral Joint Mechanics During Walking. .  37. 2019
  • Postoperative Tendon Loading With Treadmill Running Delays Tendon-to-Bone Healing: Immunohistochemical Evaluation in a Murine Rotator Cuff Repair Model. .  37. 2019
  • Effect of interposed tissue and contamination on the initial stability of a highly porous press-fit acetabular cup. .  37. 2019
  • Effect of interface mechanical discontinuities on scaffold-cartilage integration. .  37. 2019
  • Mechanical performance of cementless total knee replacements: It is not all about the maximum loads. .  37. 2019
  • Duration of postoperative immobilization affects MMP activity at the healing graft-bone interface: Evaluation in a mouse ACL reconstruction model. .  37. 2018
  • Evaluating the role of subacromial impingement in rotator cuff tendinopathy: Development and analysis of a novel murine model. .  36. 2018
  • Commentary on "increased oxidative protection by high active vitamin E content and partial radiation crosslinking of UHMWPE". 2018
  • The seating mechanics of head-neck modular tapers in vitro: Load-displacement measurements, moisture, and rate effects. .  36. 2018
  • Fixed-bearing medial unicompartmental knee arthroplasty restores neither the medial pivoting behavior nor the ligament forces of the intact knee in passive flexion. .  36. 2018
  • The onset of fretting at the head-stem connection in hip arthroplasty is affected by head material and trunnion design under simulated corrosion conditions. .  36. 2018
  • Phlpp inhibitors block pain and cartilage degradation associated with osteoarthritis. .  36. 2017
  • Influence of the pericellular and extracellular matrix structural properties on chondrocyte mechanics. .  36. 2017
  • Collagen XI mutation lowers susceptibility to load-induced cartilage damage in mice. .  36. 2017
  • Neutral glenoid alignment in reverse shoulder arthroplasty does not guarantee decreased risk of impingement. .  36. 2017
  • Correlation between RUST assessments of fracture healing to structural and biomechanical properties. .  36. 2017
  • Varus knee osteoarthritis: Elevated synovial CD15 counts correlate with inferior biomechanical properties of lateral-compartment cartilage. .  36. 2017
  • Tissue-engineered tendon constructs for rotator cuff repair in sheep. .  36. 2017
  • Substrain-specific differences in bone parameters, alpha-2-macroglobulin circulating levels, and osteonecrosis incidence in a rat model. .  35. 2017
  • Bone robusticity in two distinct skeletal dysplasias diverges from established patterns. .  35. 2017
  • Reducing uncertainty when using knee-specific finite element models by assessing the effect of input parameters. .  35. 2017
  • Standardized fluoroscopy-based technique to measure intraoperative cup anteversion. .  35. 2017
  • Pharmacological inhibition of myostatin protects against skeletal muscle atrophy and weakness after anterior cruciate ligament tear. .  35. 2017
  • Clinical platform for understanding the relationship between joint contact mechanics and articular cartilage changes after meniscal surgery. .  35. 2017
  • Kinematics of meniscal- and ACL-transected mouse knees during controlled tibial compressive loading captured using roentgen stereophotogrammetry. .  35. 2016
  • Composite metric R 2  - R 1ρ (1/T 2  - 1/T 1ρ ) as a potential MR imaging biomarker associated with changes in pain after ACL reconstruction: A six-month follow-up. .  35. 2016
  • Perlecan is required for the chondrogenic differentiation of synovial mesenchymal cells through regulation of Sox9 gene expression. .  35. 2016
  • Cadaveric gait simulation reproduces foot and ankle kinematics from population-specific inputs. .  34. 2016
  • Baseline cartilage quality is associated with voxel-based T 1ρ and T 2 following ACL reconstruction: A multicenter pilot study. .  35. 2016
  • Progressive cell-mediated changes in articular cartilage and bone in mice are initiated by a single session of controlled cyclic compressive loading. .  34. 2016
  • Use of novel chitosan hydrogels for chemical tissue bonding of autologous chondral transplants. .  34. 2016
  • Hypoxic culture conditions induce increased metabolic rate and collagen gene expression in ACL-derived cells. .  34. 2015
  • Development of a bovine decellularized extracellular matrix-biomaterial for nucleus pulposus regeneration. .  34. 2015
  • Indian hedgehog signaling and the role of graft tension in tendon-to-bone healing: Evaluation in a rat ACL reconstruction model. .  34. 2015
  • The effect of a low radiation CT protocol on accuracy of CT guided implant migration measurement: A cadaver study. .  34. 2015
  • Use of a new model allowing controlled uniaxial loading to evaluate tendon healing in a bone tunnel. .  34. 2015
  • Patterns of strain and the determination of the safe arc of motion after subscapularis repair--A biomechanical study. .  34. 2015
  • OARSI osteoarthritis cartilage histopathology assessment system: A biomechanical evaluation in the human knee. .  34. 2015
  • Novel measure of articular instability based on contact stress confirms that the anterior cruciate ligament is a critical stabilizer of the lateral compartment. .  34. 2015
  • Adverse local tissue reaction (ALTR) associated with corrosion products in metal-on-metal and dual modular neck total hip replacements is associated with upregulation of interferon gamma-mediated chemokine signaling. .  33. 2015
  • The effect of wrist surgery on the kinematic consistency of joint axis reconstruction in a static posture. .  33. 2015
  • Effects of steroids on thrombogenic markers in patients undergoing unilateral total knee arthroplasty: a prospective, double-blind, randomized controlled trial. .  33. 2015
  • Characterization of bone perfusion by dynamic contrast-enhanced magnetic resonance imaging and positron emission tomography in the Dunkin-Hartley guinea pig model of advanced osteoarthritis. .  33. 2014
  • Intermittent PTH administration and mechanical loading are anabolic for periprosthetic cancellous bone. .  33. 2014
  • Population average T2 MRI maps reveal quantitative regional transformations in the degenerating rabbit intervertebral disc that vary by lumbar level. .  33. 2014
  • Enhancement of Achilles tendon repair mediated by matrix metalloproteinase inhibition via systemic administration of doxycycline. .  32. 2013
  • Capitellum excision: Mechanical implications and clinical consequences. .  32. 2013
  • Lack of ADAM10 in endothelial cells affects osteoclasts at the chondro-osseus junction. .  32. 2013
  • Comparison of the effect of intra-tendon applications of recombinant human platelet-derived growth factor-BB, platelet-rich plasma, steroids in a rat achilles tendon collagenase model. .  32. 2013
  • Collagen scaffold meniscus implant integration in a canine model: a histological analysis. .  31. 2013
  • Trabecular bone adaptation to loading in a rabbit model is not magnitude-dependent. .  31. 2013
  • Dose-response effect of an intra-tendon application of recombinant human platelet-derived growth factor-BB (rhPDGF-BB) in a rat Achilles tendinopathy model. .  31. 2012
  • Orthopedic wear debris mediated inflammatory osteolysis is mediated in part by NALP3 inflammasome activation. .  31. 2012
  • Prognosis and predictors of ACL reconstructions using the MOON cohort: a model for comparative effectiveness studies. .  31. 2012
  • Rotator cuff tear reduces muscle fiber specific force production and induces macrophage accumulation and autophagy. .  30. 2012
  • Epidemiology and risk factors for perioperative mortality after total hip and knee arthroplasty. .  30. 2012
  • The effect of rhPTH on the healing of tendon to bone in a rat model. .  30. 2011
  • Lubricin distribution in the torn human anterior cruciate ligament and meniscus. .  29. 2011
  • ACL reconstruction using bone-tendon-bone graft engineered from the semitendinosus tendon by injection of recombinant BMP-2 in a rabbit model. .  29. 2011
  • Interleukin-1β stimulates stromal-derived factor-1α expression in human subacromial bursa. .  29. 2011
  • Diabetes mellitus alters the mechanical properties of the native tendon in an experimental rat model. .  29. 2011
  • Analysis of frictional behavior and changes in morphology resulting from cartilage articulation with porous polyurethane foams. .  28. 2010
  • Activation of MKK3/6, SAPK, and ATF-2/c-jun in ACL fibroblasts grown in 3 dimension collagen gels in response to application of cyclic strain. .  29. 2010
  • An injectable composite material containing bone morphogenetic protein-2 shortens the period of distraction osteogenesis in vivo. .  29. 2010
  • Improving bone density at the rotator cuff footprint increases supraspinatus tendon failure stress in a rat model. .  28. 2010
  • A novel locus for adolescent idiopathic scoliosis on chromosome 12p. .  27. 2009
  • Understanding patellofemoral pain with maltracking in the presence of joint laxity: complete 3D in vivo patellofemoral and tibiofemoral kinematics. .  27. 2009
  • The effect of paraformaldehyde fixation on the delayed gadolinium-enhanced MRI of cartilage (dGEMRIC) measurement. .  27. 2009
  • Tumor interstitial fluid pressure may regulate angiogenic factors in osteosarcoma. .  26. 2008
  • The effects of RANKL inhibition on fracture healing and bone strength in a mouse model of osteogenesis imperfecta. .  26. 2008
  • Expression profiling reveals alternative macrophage activation and impaired osteogenesis in periprosthetic osteolysis. .  26. 2008
  • Role of parathyroid hormone in the mechanosensitivity of fracture healing. .  25. 2007
  • Effect of crosslinking, remelting, and aging on UHMWPE damage in a linear experimental wear model. .  25. 2007
  • Relationship between cartilage stiffness and dGEMRIC index: correlation and prediction. .  25. 2007
  • Effects of distal radius malunion on distal radioulnar joint mechanics--an in vivo study. .  25. 2007
  • Release of gentamicin from a tricalcium phosphate bone implant. .  25. 2007
  • Involvement of complement receptor 3 (CR3) and scavenger receptor in macrophage responses to wear debris. .  24. 2006
  • Restoration of growth plate function following radiotherapy is driven by increased proliferative and synthetic activity of expansions of chondrocytic clones. .  24. 2006
  • In vivo cyclic axial compression affects bone healing in the mouse tibia. .  24. 2006
  • Stromal cell-derived factor 1 (SDF-1, CXCL12) is increased in subacromial bursitis and downregulated by steroid and nonsteroidal anti-inflammatory agents. .  24. 2006
  • Retrieval, experimental, and computational assessment of the performance of total knee replacements. .  24. 2006
  • Canine ACL fibroblast integrin expression and cell alignment in response to cyclic tensile strain in three-dimensional collagen gels. .  24. 2006
  • Effect of cyclic strain and plating matrix on cell proliferation and integrin expression by ligament fibroblasts. .  24. 2006
  • Molecular signaling in intervertebral disk development. .  23. 2005
  • Macrophages accumulate in the early phase of tendon-bone healing. .  23. 2005
  • The effect of ankle ligament damage and surgical reconstructions on the mechanics of the ankle and subtalar joints revealed by three-dimensional stress MRI. .  23. 2005
  • Magnetic resonance imaging in the evaluation of periprosthetic acetabular osteolysis: a cadaveric study. .  23. 2005
  • Recruitment of osteoclast precursors by stromal cell derived factor-1 (SDF-1) in giant cell tumor of bone. .  23. 2005
  • Thrombin peptide (TP508) promotes fracture repair by up-regulating inflammatory mediators, early growth factors, and increasing angiogenesis. .  23. 2004
  • A silencer element in the cartilage oligomeric matrix protein gene regulates chondrocyte-specific expression. .  22. 2004
  • Osteoblast-like cell adhesion to bone sialoprotein peptides. .  22. 2004
  • Time, stress, and location dependent chondrocyte death and collagen damage in cyclically loaded articular cartilage. .  21. 2003
  • Acceleration of cartilage repair by genetically modified chondrocytes over expressing bone morphogenetic protein-7. .  21. 2003
  • In vivo load sharing among the quadriceps components. .  21. 2003
  • J-integral fracture toughness and tearing modulus measurement of radiation cross-linked UHMWPE. .  20. 2002
  • Biosynthetic response and mechanical properties of articular cartilage after injurious compression. .  19. 2001
  • Enhanced matrix synthesis and in vitro formation of cartilage-like tissue by genetically modified chondrocytes expressing BMP-7. .  19. 2001
  • Tissue reactions to particles of bone-substitute materials in intraosseous and heterotopic sites in rats: discrimination of osteoinduction, osteocompatibility, and inflammation. .  19. 2001
  • Detection of c-fos expression in benign and malignant musculoskeletal lesions. .  19. 2001
  • Insulin-like growth factor-I is expressed by avian flexor tendon cells. .  18. 2000
  • Chondrocyte-specific enhancer regions in the COMP gene. .  18. 2000
  • Inhibition of caspase-3-like activity prevents apoptosis while retaining functionality of human chondrocytes in vitro. .  18. 2000
  • Differential expression of integrin subunits in canine knee ligament fibroblasts. .  17. 1999
  • Writing for the Journal of Orthopaedic Research. .  17. 1999
  • Characterization of chemotactic migration and growth kinetics of canine knee ligament fibroblasts. .  17. 1999
  • Intravertebral body reconstruction with an injectable in situ-setting carbonated apatite: biomechanical evaluation of a minimally invasive technique. .  17. 1999
  • Noninvasive technique for measuring in vivo three-dimensional carpal bone kinematics. .  17. 1999
  • Publication rate of abstracts presented at the annual meeting of the Orthopaedic Research Society. .  16. 1998
  • Chemotherapy and P-glycoprotein expression in chondrosarcoma. .  16. 1998
  • Identification of the oim mutation by dye terminator chemistry combined with automated direct DNA sequencing. .  16. 1998
  • Residual stresses in ultra-high molecular weight polyethylene loaded cyclically by a rigid moving indenter in nonconforming geometries. .  16. 1998
  • Immunolocalization of cytokines and their receptors in adhesive capsulitis of the shoulder. .  15. 1997
  • Optimal normalization tests for shoulder muscle activation: an electromyographic study. .  14. 1996
  • Primary immunolocalization of estrogen and progesterone target cells in the human anterior cruciate ligament. .  14. 1996
  • Expression of interleukin-6 in osteoarthritic chondrocytes and effects of fluid-induced shear on this expression in normal human chondrocytes in vitro. .  14. 1996
  • Qualitative and quantitative analysis of orthotopic bone regeneration by marrow. .  14. 1996
  • Effect of stress deprivation and cyclic tensile loading on the material and morphologic properties of canine flexor digitorum profundus tendon: an in vitro study. .  13. 1995
  • Effects of fluid-induced shear on articular chondrocyte morphology and metabolism in vitro. .  13. 1995
  • Osteosarcoma oncogene expression detected by in situ hybridization. .  13. 1995
  • Contact areas in the thumb carpometacarpal joint. .  13. 1995
  • Immunolocalization and expression of bone morphogenetic proteins 2 and 4 in fracture healing. .  13. 1995
  • Clinical reproducibility of dual energy x-ray absorptiometry. .  13. 1995
  • Cyclic compressive loading results in fatigue cracks in ultra high molecular weight polyethylene. .  13. 1995
  • Diminished material properties and altered bone structure in rat femora during pregnancy. .  13. 1995
  • Mechanical and biochemical changes in the superficial zone of articular cartilage in canine experimental osteoarthritis. .  12. 1994
  • Effect of cyclic and static tensile loading on water content and solute diffusion in canine flexor tendons: an in vitro study. .  12. 1994
  • Treatment of experimental osteomyelitis with antibiotic-impregnated bone graft substitute. .  11. 1993
  • Evaluation of bone-grafting materials in a new canine segmental spinal fusion model. .  11. 1993
  • Effects of prefracture irradiation on the biomechanical parameters of fracture healing. .  11. 1993
  • In vivo effects of naproxen on composition, proteoglycan metabolism, and matrix metalloproteinase activities in canine articular cartilage. .  11. 1993
  • The inositol phosphate pathway as a mediator in the proliferative response of rat calvarial bone cells to cyclical biaxial mechanical strain. .  10. 1992
  • Effects of propranolol on bone metabolism in the rat. .  9. 1991
  • Mechanical characteristics of the stem-cement interface. .  9. 1991
  • Measurement reproducibility of two commercial knee test devices. .  9. 1991
  • Interspecies comparisons of in situ intrinsic mechanical properties of distal femoral cartilage. .  9. 1991
  • Purified staphylococcal culture medium stimulates neutral metalloprotease secretion from human articular cartilage. .  9. 1991
  • An investigation of a compliant interface for press-fit joint replacement. .  8. 1990
  • Omental angiogenic lipid fraction and bone repair. An experimental study in the rat. .  7. 1989
  • Prevention of ligament and meniscus atrophy by active joint motion in a non-weight-bearing model. .  7. 1989
  • The repair of experimentally produced defects in rabbit articular cartilage by autologous chondrocyte transplantation. .  7. 1989
  • Biomechanics of the craniocervical region: the alar and transverse ligaments. .  6. 1988
  • The effect of cryopreservation on canine menisci: a biochemical, morphologic, and biomechanical evaluation. .  6. 1988
  • Water content and equilibrium water partition in immature cartilage. .  6. 1988
  • Biochemical characterization of fracture callus proteoglycans. .  5. 1987
  • Valgus stability of the elbow. .  5. 1987
  • Immunofluorescent localization of structural collagen types in endochondral fracture repair. .  4. 1986
  • Influence of cartilage conformation on its equilibrium water partition. .  3. 1985
  • Fatigue crack propagation behavior of ultrahigh molecular weight polyethylene. .  2. 1984
  • RE: Talks BJ, Fernquest S, Palmer A, et al. 2019. No Evidence of Systemic Inflammation in Symptomatic Patients With Femoroacetabular Impingement. .  37. 2019
  • Commentary on "Third-generation autologous chondrocyte implantation versus mosaicplasty for knee cartilage injury: 2-year randomized trial". .  34. 2016
  • Research involving ultra high molecular weight polyethylene. .  13. 1995
  • Quadriceps protects the anterior cruciate ligament. 2002
  • Irrigation and debridement. 2014
  • Prevention of late PJI. 2014
  • Plagiarism: an assault on the integrity of scientific research. .  30. 2012
  • A new feature of the journal of orthopaedic research: research perspectives. .  29. 2011
  • Orthopaedic surgeons, scientists, and industry. .  26. 2008
  • Patient care, professionalism, and relations with industry. .  26. 2008
  • Celebrating 25 years of Advancing Orthopaedic Research. .  26. 2008
  • Guidelines for describing mouse skeletal phenotype. .  21. 2003
  • Scientific progress and dissemination of research results: abstracts versus peer-reviewed publications. .  16. 1998
  • Scientific credibility requires complete presentation of methods. .  15. 1997
  • Meniscal repair: The current state and recent advances in augmentation. 2021
  • The role of the macrophage in tendinopathy and tendon healing. 2020
  • Biological and Mechanical Predictors of Meniscus Function: Basic Science to Clinical Translation. 2019
  • Integrating soft and hard tissues via interface tissue engineering. 2018
  • Review of current understanding of post-traumatic osteoarthritis resulting from sports injuries. 2016
  • Morphologic and quantitative magnetic resonance imaging of knee articular cartilage for the assessment of post-traumatic osteoarthritis. 2016
  • Reproducibility in modeling and simulation of the knee: Academic, industry, and regulatory perspectives. 2023
  • Clinical magnetic resonance imaging of arthroplasty at 1.5 T. 2020
  • The biologic effects of implant materials. 1991

ISO Abbreviation

  • J Orthop Res

Linking ISSN

International standard serial number (issn), electronic international standard serial number (eissn).

Journal of Orthopaedic Surgery and Research

Collection image

Safer 3D-printed personalised orthopaedic implants

Editor: Peter Vee Sin Lee:  University of Melbourne, Australia Submission Status: Open until 30 April 2024

Aims and Scope

Journal of Orthopaedic Surgery and Research  is an open access, peer-reviewed online journal that encompasses all aspects of clinical and basic research studies related to musculoskeletal issues.

Journal of Orthopaedic Surgery and Research  provides the platform for exchange of new clinical and scientific information in the most precise and expeditious way to achieve timely dissemination of information and cross-fertilization of ideas.

Please note: Journal of Orthopaedic Surgery and Research  does not consider or publish case reports. Review the submission guidelines for details of which article types are published in the journal. 

Featured Article: Vacuum-assisted closure versus conventional dressing in necrotizing fasciitis: a systematic review and meta-analysis

New Content Item

  • Most accessed

Investigating the reliability and validity of subacromial space measurements using ultrasound and MRI

Authors: Stephen M. Boulanger, Alexandra Mahna, Talia Alenabi, Anthony A. Gatti, Oriana Culig, Loriann M. Hynes and Jaclyn N. Chopp-Hurley

Caulis Spatholobi extracts inhibit osteosarcoma growth and metastasis through suppression of CXCR4/PI3K/AKT signaling

Authors: Yang Jiang, Yemei Gao, Xin Li, Fangming He, Yang Liu and Renxian Wang

Randomized controlled trial of vitamin d supplementation on toll-like receptor-2 (tlr-2) and toll-like receptor-4 (tlr-4) in tuberculosis spondylitis patients

Authors: Jainal Arifin, Muhammad Nasrum Massi, Karya Triko Biakto, Agussalim Bukhari, Zairin Noor and Muhammad Phetrus Johan

3D-printed PCL/β-TCP/CS composite artificial bone and histocompatibility study

Authors: Chao Zheng and Mingman Zhang

Sex-based differences in clinical and radiological presentation of patients with degenerative lumbar scoliosis: a cross-sectional study

Authors: Xiao Liang, Puxin Yang, Hongru Yuan, Yachong Huo, Dalong Yang, Hui Wang and Wenyuan Ding

Most recent articles RSS

View all articles

Extracorporeal shockwave therapy in musculoskeletal disorders

Authors: Ching-Jen Wang

Pedicle screw loosening: the value of radiological imagings and the identification of risk factors assessed by extraction torque during screw removal surgery

Authors: Xiaoliang Wu, Jiawei Shi, Jinyan Wu, Yongquan Cheng, Kaiwen Peng, Jianting Chen and Hui Jiang

Prophylaxis of heterotopic ossification – an updated review

Authors: Evan O Baird and Qian K Kang

The global prevalence of osteoporosis in the world: a comprehensive systematic review and meta-analysis

Authors: Nader Salari, Hooman Ghasemi, Loghman Mohammadi, Mohammad hasan Behzadi, Elham Rabieenia, Shamarina Shohaimi and Masoud Mohammadi

Sports-related wrist and hand injuries: a review

Authors: Daniel M. Avery III, Craig M. Rodner and Cory M. Edgar

Most accessed articles RSS

Recruiting Associate Editors

Journal of Orthopaedic Surgery and Research invites you to join our team of expert Associate Editors in handling and assessing manuscripts. We are looking for editors with experience in the fields of spinal surgery, basic sciences and genetics in orthopaedics. For consideration, please send your CV and details of any editorial experience to [email protected]

Trending articles

altmetric (1)

Important Information for Authors: In Review

Journal of Orthopaedic Surgery and Research now participates in In Review. Learn more here

Overcoming Language Barriers

English Editing discount

Latest Tweets

Your browser needs to have JavaScript enabled to view this timeline

Editors' profiles


writing for the journal of orthopaedic research

Prof Peter V. Lee holds a full Professor position at the University of Melbourne since 2015. He has held several leadership positions including Deputy Head and Acting Head of Department of Mechanical Engineering, and currently as the Deputy Head (Research and Engagement) of Department of Biomedical Engineering, and the Melbourne School of Engineering Research Program Leader for Medical and Pharmaceutical Technologies. He also leads the Cell and Tissue Biomechanics Laboratory in Engineering, where his research aims to better understand the behaviour of biological cells and tissues under mechanical forces.

JOSR logo

  • Editorial Board
  • Manuscript editing services
  • Instructions for Editors
  • Sign up for article alerts and news from this journal
  • Follow us on Twitter

Annual Journal Metrics

2022 Citation Impact 2.6 - 2-year Impact Factor 3.0 - 5-year Impact Factor 1.251 - SNIP (Source Normalized Impact per Paper) 0.744 - SJR (SCImago Journal Rank)

2022 Speed 4 days submission to first editorial decision for all manuscripts (Median) 81 days submission to accept (Median)

2022 Usage  2,399,834 downloads 1,469 Altmetric mentions

  • More about our metrics

ISSN: 1749-799X

  • Latest Headlines
  • English Edition Edition English 中文 (Chinese) 日本語 (Japanese)
  • Print Edition
  • More More Other Products from WSJ Buy Side from WSJ WSJ Shop WSJ Wine

This copy is for your personal, non-commercial use only. Distribution and use of this material are governed by our Subscriber Agreement and by copyright law. For non-personal use or to order multiple copies, please contact Dow Jones Reprints at 1-800-843-0008 or visit www.djreprints.com.


Claudine Gay and Why Academic Honesty Matters

Politicization and lower standards jeopardize harvard’s standing as a great research institution..

James Hankins

Dec. 27, 2023 5:36 pm ET


Claudine Gay , the president of my university, is under attack for academic dishonesty. She is charged with several instances of plagiarism, in her dissertation and other published work, in addition to data falsification. As of this writing it seems not unlikely that she may be fired or asked to resign.

Copyright © 2024 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8

  • Top Resume : Top Resume Coupon: 10% Off professional resume writing
  • Walmart : $20 Off Walmart Promo Code - Any $50+ Order
  • Groupon : Groupon New Year coupon: Up to 75% off all local deals + extra 30% off at checkout
  • eBay : Unlock 10% Off orders With this eBay coupon
  • AliExpress : $3 Off using AliExpress New Users Coupon
  • JCPenney : Get 25% off your Online Purchases w/ JCPenney Coupon Code

Most Popular news

Most popular opinion, most popular opinion, recommended videos.

Copyright © 2024 Dow Jones & Company, Inc. All Rights Reserved

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • View all journals
  • My Account Login
  • Explore content
  • About the journal
  • Publish with us
  • Sign up for alerts
  • Open access
  • Published: 20 December 2023

Autonomous chemical research with large language models

  • Daniil A. Boiko   ORCID: orcid.org/0000-0003-4140-4645 1 ,
  • Robert MacKnight 1 ,
  • Ben Kline 2 &
  • Gabe Gomes   ORCID: orcid.org/0000-0002-8235-5969 1 , 3 , 4  

Nature volume  624 ,  pages 570–578 ( 2023 ) Cite this article

57k Accesses

2 Citations

881 Altmetric

Metrics details

  • Computer science

Transformer-based large language models are making significant strides in various fields, such as natural language processing 1 , 2 , 3 , 4 , 5 , biology 6 , 7 , chemistry 8 , 9 , 10 and computer programming 11 , 12 . Here, we show the development and capabilities of Coscientist, an artificial intelligence system driven by GPT-4 that autonomously designs, plans and performs complex experiments by incorporating large language models empowered by tools such as internet and documentation search, code execution and experimental automation. Coscientist showcases its potential for accelerating research across six diverse tasks, including the successful reaction optimization of palladium-catalysed cross-couplings, while exhibiting advanced capabilities for (semi-)autonomous experimental design and execution. Our findings demonstrate the versatility, efficacy and explainability of artificial intelligence systems like Coscientist in advancing research.

Large language models (LLMs), particularly transformer-based models, are experiencing rapid advancements in recent years. These models have been successfully applied to various domains, including natural language 1 , 2 , 3 , 4 , 5 , biological 6 , 7 and chemical research 8 , 9 , 10 as well as code generation 11 , 12 . Extreme scaling of models 13 , as demonstrated by OpenAI, has led to significant breakthroughs in the field 1 , 14 . Moreover, techniques such as reinforcement learning from human feedback 15 can considerably enhance the quality of generated text and the models’ capability to perform diverse tasks while reasoning about their decisions 16 .

On 14 March 2023, OpenAI released their most capable LLM to date, GPT-4 14 . Although specific details about the model training, sizes and data used are limited in GPT-4’s technical report, OpenAI researchers have provided substantial evidence of the model’s exceptional problem-solving abilities. Those include—but are not limited to—high percentiles on the SAT and BAR examinations, LeetCode challenges and contextual explanations from images, including niche jokes 14 . Moreover, the technical report provides an example of how the model can be used to address chemistry-related problems.

Simultaneously, substantial progress has been made toward the automation of chemical research. Examples range from the autonomous discovery 17 , 18 and optimization of organic reactions 19 to the development of automated flow systems 20 , 21 and mobile platforms 22 .

The combination of laboratory automation technologies with powerful LLMs opens the door to the development of a sought-after system that autonomously designs and executes scientific experiments. To accomplish this, we intended to address the following questions. What are the capabilities of LLMs in the scientific process? What degree of autonomy can we achieve? How can we understand the decisions made by autonomous agents?

In this work, we present a multi-LLMs-based intelligent agent (hereafter simply called Coscientist) capable of autonomous design, planning and performance of complex scientific experiments. Coscientist can use tools to browse the internet and relevant documentation, use robotic experimentation application programming interfaces (APIs) and leverage other LLMs for various tasks. This work has been done independently and in parallel to other works on autonomous agents 23 , 24 , 25 , with ChemCrow 26 serving as another example in the chemistry domain. In this paper, we demonstrate the versatility and performance of Coscientist in six tasks: (1) planning chemical syntheses of known compounds using publicly available data; (2) efficiently searching and navigating through extensive hardware documentation; (3) using documentation to execute high-level commands in a cloud laboratory; (4) precisely controlling liquid handling instruments with low-level instructions; (5) tackling complex scientific tasks that demand simultaneous use of multiple hardware modules and integration of diverse data sources; and (6) solving optimization problems requiring analyses of previously collected experimental data.

Coscientist system architecture

Coscientist acquires the necessary knowledge to solve a complex problem by interacting with multiple modules (web and documentation search, code execution) and by performing experiments. The main module (‘Planner’) has the goal of planning, based on the user input by invoking the commands defined below. The Planner is a GPT-4 chat completion instance serving the role of an assistant. The initial user input along with command outputs are treated as user messages to the Planner. System prompts (static inputs defining the LLMs’ goals) for the Planner are engineered 1 , 27 in a modular fashion, described as four commands that define the action space: ‘GOOGLE’, ‘PYTHON’, ‘DOCUMENTATION’ and ‘EXPERIMENT’. The Planner calls on each of these commands as needed to collect knowledge. The GOOGLE command is responsible for searching the internet with the ‘Web searcher’ module, which is another LLM itself. The PYTHON command allows the Planner to perform calculations to prepare the experiment using a ‘Code execution’ module. The EXPERIMENT command actualizes ‘Automation’ through APIs described by the DOCUMENTATION module. Like GOOGLE, the DOCUMENTATION command provides information to the main module from a source, in this case documentation concerning the desired API. In this study, we have demonstrated the compatibility with the Opentrons Python API and the Emerald Cloud Lab (ECL) Symbolic Lab Language (SLL). Together, these modules make up Coscientist, which receives a simple plain text input prompt from the user (for example, “perform multiple Suzuki reactions”). This architecture is depicted in Fig. 1 .

figure 1

a , Coscientist is composed of multiple modules that exchange messages. Boxes with blue background represent LLM modules, the Planner module is shown in green, and the input prompt is in red. White boxes represent modules that do not use LLMs. b , Types of experiments performed to demonstrate the capabilities when using individual modules or their combinations. c , Image of the experimental setup with a liquid handler. UV-Vis, ultraviolet visible.

Furthermore, some of the commands can use subactions. The GOOGLE command is capable of transforming prompts into appropriate web search queries, running them against the Google Search API, browsing web pages and funneling answers back to the Planner. Similarly, the DOCUMENTATION command performs retrieval and summarization of necessary documentation (for example, robotic liquid handler or a cloud laboratory) for Planner to invoke the EXPERIMENT command.

The PYTHON command performs code execution (not reliant upon any language model) using an isolated Docker container to protect the users’ machine from any unexpected actions requested by the Planner. Importantly, the language model behind the Planner enables code to be fixed in case of software errors. The same applies to the EXPERIMENT command of the Automation module, which executes generated code on corresponding hardware or provides the synthetic procedure for manual experimentation.

Web search module

To demonstrate one of the functionalities of the Web Searcher module, we designed a test set composed of seven compounds to synthesize, as presented in Fig. 2a . The Web Searcher module versions are represented as ‘search-gpt-4’ and ‘search-gpt-3.5-turbo’. Our baselines include OpenAI’s GPT-3.5 and GPT-4, Anthropic’s Claude 1.3 28 and Falcon-40B-Instruct 29 —considered one of the best open-source models at the time of this experiment as per the OpenLLM leaderboard 30 .

figure 2

a , Comparison of various LLMs on compound synthesis benchmarks. Error bars represent s.d. values. b , Two examples of generated syntheses of nitroaniline. c , Two example of generated syntheses of ibuprofen. UV, ultraviolet.

We prompted every model to provide a detailed compound synthesis, ranking the outputs on the following scale (Fig. 2 ):

5 for a very detailed and chemically accurate procedure description

4 for a detailed and chemically accurate description but without reagent quantities

3 for a correct chemistry description that does not include step-by-step procedure

2 for extremely vague or unfeasible descriptions

1 for incorrect responses or failure to follow instructions

All scores below 3 indicate task failure. It is important to note that all answers between 3 and 5 are chemically correct but offer varying levels of detail. Despite our attempts to better formalize the scale, labelling is inherently subjective and so, may be different between the labelers.

Across non-browsing models, the two versions of the GPT-4 model performed best, with Claude v.1.3 demonstrating similar performance. GPT-3 performed significantly worse, and Falcon 40B failed in most cases. All non-browsing models incorrectly synthesized ibuprofen (Fig. 2c ). Nitroaniline is another example; although some generalization of chemical knowledge might inspire the model to propose direct nitration, this approach is not experimentally applicable as it would produce a mixture of compounds with a very minor amount of the product (Fig. 2b ). Only the GPT-4 models occasionally provided the correct answer.

The GPT-4-powered Web Searcher significantly improves on synthesis planning. It reached maximum scores across all trials for acetaminophen, aspirin, nitroaniline and phenolphthalein (Fig. 2b ). Although it was the only one to achieve the minimum acceptable score of three for ibuprofen, it performed lower than some of the other models for ethylacetate and benzoic acid, possibly because of the widespread nature of these compounds. These results show the importance of grounding LLMs to avoid ‘hallucinations’ 31 . Overall, the performance of GPT-3.5-enabled Web Searcher trailed its GPT-4 competition, mainly because of its failure to follow specific instructions regarding output format.

Extending the Planner’s action space to leverage reaction databases, such as Reaxys 32 or SciFinder 33 , should significantly enhance the system’s performance (especially for multistep syntheses). Alternatively, analysing the system’s previous statements is another approach to improving its accuracy. This can be done through advanced prompting strategies, such as ReAct 34 , Chain of Thought 35 and Tree of Thoughts 36 .

Documentation search module

Addressing the complexities of software components and their interactions is crucial for integrating LLMs with laboratory automation. A key challenge lies in enabling Coscientist to effectively utilize technical documentation. LLMs can refine their understanding of common APIs, such as the Opentrons Python API 37 , by interpreting and learning from relevant technical documentation. Furthermore, we show how GPT-4 can learn how to programme in the ECL SLL.

Our approach involved equipping Coscientist with essential documentation tailored to specific tasks (as illustrated in Fig. 3a ), allowing it to refine its accuracy in using the API and improve its performance in automating experiments.

figure 3

a , Prompt-to-code through ada embedding and distance-based vector search. b , Example of code for using OT-2’s heater–shaker module. c , Prompt-to-function/prompt-to-SLL (to symbolic laboratory language) through supplementation of documentation. d , Example of valid ECL SLL code for performing high-performance liquid chromatography (HPLC) experiments.

Information retrieval systems are usually based on two candidate selection approaches: inverted search index and vector database 38 , 39 , 40 , 41 . For the first one, each unique word in the search index is mapped to the documents containing it. At inference time, all documents containing words from a query are selected and ranked based on various manually defined formulas 42 . The second approach starts by embedding the documents with neural networks or as term frequency–inverse document frequency embedding vectors 43 , followed by the construction of a vector database. Retrieval of similar vectors from this database occurs at inference time, usually using one of the approximate nearest neighbour search algorithms 44 . When strategies such as Transformer models are used, there are more chances to account for synonyms natively without doing synonym-based query expansion, as would be done in the first approach 45 .

Following the second approach, all sections of the OT-2 API documentation were embedded using OpenAI’s ada model. To ensure proper use of the API, an ada embedding for the Planner’s query was generated, and documentation sections are selected through a distance-based vector search. This approach proved critical for providing Coscientist with information about the heater–shaker hardware module necessary for performing chemical reactions (Fig. 3b ).

A greater challenge emerges when applying this approach to a more diverse robotic ecosystem, such as the ECL. Nonetheless, we can explore the effectiveness of providing information about the ECL SLL, which is currently unknown to the GPT-4 model. We conducted three separate investigations concerning the SLL: (1) prompt-to-function; (2) prompt-to-SLL; and (3) prompt-to-samples. Those investigations are detailed in Supplementary Information section ‘ ECL experiments ’.

For investigation 1, we provide the Docs searcher with a documentation guide from ECL pertaining to all available functions for running experiments 46 . Figure 3c summarizes an example of the user providing a simple prompt to the system, with the Planner receiving relevant ECL functions. In all cases, functions are correctly identified for the task.

Figure 3c,d continues to describe investigation 2, the prompt-to-SLL investigation. A single appropriate function is selected for the task, and the documentation is passed through a separate GPT-4 model to perform code retention and summarization. After the complete documentation has been processed, the Planner receives usage information to provide EXPERIMENT code in the SLL. For instance, we provide a simple example that requires the ‘ExperimentHPLC’ function. Proper use of this function requires familiarity with specific ‘Models’ and ‘Objects’ as they are defined in the SLL. Generated code was successfully executed at ECL; this is available in Supplementary Information . The sample was a caffeine standard sample. Other parameters (column, mobile phases, gradients) were determined by ECL’s internal software (a high-level description is in Supplementary Information section ‘ HPLC experiment parameter estimation ’). Results of the experiment are provided in Supplementary Information section ‘ Results of the HPLC experiment in the cloud lab ’. One can see that the air bubble was injected along with the analyte’s solution. This demonstrates the importance of development of automated techniques for quality control in cloud laboratories. Follow-up experiments leveraging web search to specify and/or refine additional experimental parameters (column chemistry, buffer system, gradient and so on) would be required to optimize the experimental results. Further details on this investigation are in Supplementary Information section ‘ Analysis of ECL documentation search results ’.

A separate prompt-to-samples investigation, investigation 3, was conducted by providing a catalogue of available samples, enabling the identification of relevant stock solutions that are on ECL’s shelves. To showcase this feature, we provide the Docs searcher module with all 1,110 Model samples from the catalogue. By simply providing a search term (for example, ‘Acetonitrile’), all relevant samples are returned. This is also available in Supplementary Information .

Controlling laboratory hardware

Access to documentation enables us to provide sufficient information for Coscientist to conduct experiments in the physical world. To initiate the investigation, we chose the Opentrons OT-2, an open-source liquid handler with a well-documented Python API. The ‘Getting Started’ page from its documentation was supplied to the Planner in the system prompt. Other pages were vectorized using the approach described above. For this investigation, we did not grant access to the internet (Fig. 4a ).

figure 4

a , Overview of Coscientist’s configuration. b , Drawing a red cross. c , Colouring every other row. d , Drawing a yellow rectangle. e , Drawing a blue diagonal.

We started with simple plate layout-specific experiments. Straightforward prompts in natural language, such as “colour every other line with one colour of your choice”, resulted in accurate protocols. When executed by the robot, these protocols closely resembled the requested prompt (Fig. 4b–e ).

Ultimately, we aimed to assess the system’s ability to integrate multiple modules simultaneously. Specifically, we provided the ‘UVVIS’ command, which can be used to pass a microplate to plate reader working in the ultraviolet–visible wavelength range. To evaluate Coscientist’s capabilities to use multiple hardware tools, we designed a toy task; in 3 wells of a 96-well plate, three different colours are present—red, yellow and blue. The system must determine the colours and their positions on the plate without any prior information.

The Coscientist’s first action was to prepare small samples of the original solutions (Extended Data Fig. 1 ). Ultraviolet-visible measurements were then requested to be performed by the Coscientist (Supplementary Information section ‘ Solving the colours problem’ and Supplementary Fig. 1 ). Once completed, Coscientist was provided with a file name containing a NumPy array with spectra for each well of the microplate. Coscientist subsequently generated Python code to identify the wavelengths with maximum absorbance and used these data to correctly solve the problem, although it required a guiding prompt asking it to think through how different colours absorb light.

Integrated chemical experiment design

We evaluated Coscientist’s ability to plan catalytic cross-coupling experiments by using data from the internet, performing the necessary calculations and ultimately, writing code for the liquid handler. To increase complexity, we asked Coscientist to use the OT-2 heater–shaker module released after the GPT-4 training data collection cutoff. The available commands and actions supplied to the Coscientist are shown in Fig. 5a . Although our setup is not yet fully automated (plates were moved manually), no human decision-making was involved.

figure 5

a , Overview of Coscientist’s configuration. b , Available compounds (DMF, dimethylformamide; DiPP, 2,6-diisopropylphenyl). c , Liquid handler setup. d , Solving the synthesis problem. e , Comparison of reagent selection performance with a large dataset. f , Comparison of reagent choices across multiple runs. g , Overview of justifications made when selecting various aryl halides. h , Frequency of visited URLs. i , Total ion current (TIC) chromatogram of the Suzuki reaction mixture (top panel) and the pure standard, mass spectra at 9.53 min (middle panel) representing the expected reaction product and mass spectra of the pure standard (bottom panel). j , TIC chromatogram of the Sonogashira reaction mixture (top panel) and the pure standard, mass spectra at 12.92 min (middle panel) representing the expected reaction product and mass spectra of the pure standard (bottom panel). Rel., relative.

The test challenge for Coscientist’s complex chemical experimentation capabilities was designed as follows. (1) Coscientist is provided with a liquid handler equipped with two microplates (source and target plates). (2) The source plate contains stock solutions of multiple reagents, including phenyl acetylene and phenylboronic acid, multiple aryl halide coupling partners, two catalysts, two bases and the solvent to dissolve the sample (Fig. 5b ). (3) The target plate is installed on the OT-2 heater–shaker module (Fig. 5c ). (4) Coscientist’s goal is to successfully design and perform a protocol for Suzuki–Miyaura and Sonogashira coupling reactions given the available resources.

To start, Coscientist searches the internet for information on the requested reactions, their stoichiometries and conditions (Fig. 5d ). The correct coupling partners are selected for the corresponding reactions. Designing and performing the requested experiments, the strategy of Coscientist changes among runs (Fig. 5f ). Importantly, the system does not make chemistry mistakes (for instance, it never selects phenylboronic acid for the Sonogashira reaction). Interestingly, the base DBU (1,8-diazabicyclo[5.4.0]undec-7-ene) is selected more often with the PEPPSI–IPr (PEPPSI, pyridine-enhanced precatalyst preparation stabilization and initiation; IPr, 1,3-bis(2,6-diisopropylphenyl)imidazol-2-ylidene) complex, with that preference switching in Sonogashira reaction experiments; likewise, bromobenzene is chosen more often for Suzuki than for Sonogashira couplings. Additionally, the model can provide justifications on specific choices (Fig. 5g ), demonstrating the ability to operate with concepts such as reactivity and selectivity (more details are in Supplementary Information section ‘ Analysis of behaviour across multiple runs ’). This capability highlights a potential future use case to analyse the reasoning of the LLMs used by performing experiments multiple times. Although the Web Searcher visited various websites (Fig. 5h ), overall Coscientist retrieves Wikipedia pages in approximately half of cases; notably, American Chemical Society and Royal Society of Chemistry journals are amongst the top five sources.

Coscientist then calculates the required volumes of all reactants and writes a Python protocol for running the experiment on the OT-2 robot. However, an incorrect heater–shaker module method name was used. Upon making this mistake, Coscientist uses the Docs searcher module to consult the OT-2 documentation. Next, Coscientist modifies the protocol to a corrected version, which ran successfully (Extended Data Fig. 2 ). Subsequent gas chromatography–mass spectrometry analysis of the reaction mixtures revealed the formation of the target products for both reactions. For the Suzuki reaction, there is a signal in the chromatogram at 9.53 min where the mass spectra match the mass spectra for biphenyl (corresponding molecular ion mass-to-charge ratio and fragment at 76 Da) (Fig. 5i ). For the Sonogashira reaction, we see a signal at 12.92 min with a matching molecular ion mass-to-charge ratio; the fragmentation pattern also looks very close to the one from the spectra of the reference compound (Fig. 5j ). Details are in Supplementary Information section ‘ Results of the experimental study ’.

Although this example requires Coscientist to reason on which reagents are most suitable, our experimental capabilities at that point limited the possible compound space to be explored. To address this, we performed several computational experiments to evaluate how a similar approach can be used to retrieve compounds from large compound libraries 47 . Figure 5e shows Coscientist’s performance across five common organic transformations, with outcomes depending on the queried reaction and its specific run (the GitHub repository has more details). For each reaction, Coscientist was tasked with generating reactions for compounds from a simplified molecular-input line-entry system (SMILES) database. To achieve the task, Coscientist uses web search and code execution with the RDKit chemoinformatics package.

Chemical reasoning capabilities

The system demonstrates appreciable reasoning capabilities, enabling the request of necessary information, solving of multistep problems and generation of code for experimental design. Some researchers believe that the community is only starting to understand all the capabilities of GPT-4 (ref. 48 ). OpenAI has shown that GPT-4 could rely on some of those capabilities to take actions in the physical world during their initial red team testing performed by the Alignment Research Center 14 .

One of the possible strategies to evaluate an intelligent agent’s reasoning capabilities is to test if it can use previously collected data to guide future actions. Here, we focused on the multi-variable design and optimization of Pd-catalysed transformations, showcasing Coscientist’s abilities to tackle real-world experimental campaigns involving thousands of examples. Instead of connecting LLMs to an optimization algorithm as previously done by Ramos et al. 49 , we aimed to use Coscientist directly.

We selected two datasets containing fully mapped reaction condition spaces where yield was available for all combinations of variables. One is a Suzuki reaction dataset collected by Perera et al. 50 , where these reactions were performed in flow with varying ligands, reagents/bases and solvents (Fig. 6a ). Another is Doyle’s Buchwald–Hartwig reaction dataset 51 (Fig. 6e ), where variations in ligands, additives and bases were recorded. At this point, any reaction proposed by Coscientist would be within these datasets and accessible as a lookup table.

figure 6

a , A general reaction scheme from the flow synthesis dataset analysed in c and d . b , The mathematical expression used to calculate normalized advantage values. c , Comparison of the three approaches (GPT-4 with prior information, GPT-4 without prior information and GPT-3.5 without prior information) used to perform the optimization process. d , Derivatives of the NMA and normalized advantage values evaluated in c , left and centre panels. e , Reaction from the C–N cross-coupling dataset analysed in f and g . f , Comparison of two approaches using compound names and SMILES string as compound representations. g , Coscientist can reason about electronic properties of the compounds, even when those are represented as SMILES strings. DMSO, dimethyl sulfoxide.

We designed the Coscientist’s chemical reasoning capabilities test as a game with the goal of maximizing the reaction yield. The game’s actions consisted of selecting specific reaction conditions with a sensible chemical explanation while listing the player’s observations about the outcome of the previous iteration. The only hard rule was for the player to provide its actions written in JavaScript Object Notation (JSON) format. If the JSON file could not be parsed, the player is alerted of its failure to follow the specified data format. The player had a maximum of 20 iterations (accounting for 5.2% and 6.9% of the total space for the first and second datasets, respectively) to finish the game.

We evaluate Coscientist’s performance using the normalized advantage metric (Fig. 6b ). Advantage is defined as the difference between a given iteration yield and the average yield (advantage over a random strategy). Normalized advantage measures the ratio between advantage and maximum advantage (that is, the difference between the maximum and average yield). The normalized advantage metric has a value of one if the maximum yield is reached, zero if the system exhibits completely random behaviour and less than zero if the performance at this step is worse than random. An increase in normalized advantage over each iteration demonstrates Coscientist’s chemical reasoning capabilities. The best result for a given iteration can be evaluated using the normalized maximum advantage (NMA), which is the normalized value of the maximum advantage achieved until the current step. As NMA cannot decrease, the valuable observations come in the form of the rate of its increase and its final point. Finally, during the first step, the values for NMA and normalized advantage equal each other, portraying the model’s prior knowledge (or lack thereof) without any data being collected.

For the Suzuki dataset, we compared three separate approaches: (1) GPT-4 with prior information included in the prompt (which consisted of 10 yields from random combinations of reagents); (2) GPT-4; or (3) GPT-3.5 without any prior information (Fig. 6c ). When comparing GPT-4 with the inclusion and exclusion of prior information, it is clear that the initial guess for the former scenario is better, which aligns with our expectations considering the provided information about the system’s reactivity. Notably, when excluding prior information, there are some poor initial guesses, whereas there are none when the model has prior information. However, at the limit, the models converge to the same NMA. The GPT-3.5 model plots have a very limited number of data points, primarily because of its inability to output messages in the correct JSON schema as requested in the prompt. It is unclear if the GPT-4 training data contain any information from these datasets. If so, one would expect that the initial model guess would be better than what we observed.

The normalized advantage values increase over time, suggesting that the model can effectively reuse the information obtained to provide more specific guidance on reactivity. Evaluating the derivative plots (Fig. 6d ) does not show any significant difference between instances with and without the input of prior information.

There are many established optimization algorithms for chemical reactions. In comparison with standard Bayesian optimization 52 , both GPT-4-based approaches show higher NMA and normalized advantage values (Fig. 6c ). A detailed overview of the exact Bayesian optimization strategy used is provided in Supplementary Information section ‘ Bayesian optimization procedure’ . It is observed that Bayesian optimization’s normalized advantage line stays around zero and does not increase over time. This may be caused by different exploration/exploitation balance for these two approaches and may not be indicative of their performance. For this purpose, the NMA plot should be used. Changing the number of initial samples does not improve the Bayesian optimization trajectory (Extended Data Fig. 3a ). Finally, this performance trend is observed for each unique substrate pairings (Extended Data Fig. 3b ).

For the Buchwald–Hartwig dataset (Fig. 6e ), we compared a version of GPT-4 without prior information operating over compound names or over compound SMILES strings. It is evident that both instances have very similar performance levels (Fig. 6f ). However, in certain scenarios, the model demonstrates the ability to reason about the reactivity of these compounds simply by being provided their SMILES strings (Fig. 6g ).

In this paper, we presented a proof of concept for an artificial intelligent agent system capable of (semi-)autonomously designing, planning and multistep executing scientific experiments. Our system demonstrates advanced reasoning and experimental design capabilities, addressing complex scientific problems and generating high-quality code. These capabilities emerge when LLMs gain access to relevant research tools, such as internet and documentation search, coding environments and robotic experimentation platforms. The development of more integrated scientific tools for LLMs has potential to greatly accelerate new discoveries.

The development of new intelligent agent systems and automated methods for conducting scientific experiments raises potential concerns about the safety and potential dual-use consequences, particularly in relation to the proliferation of illicit activities and security threats. By ensuring the ethical and responsible use of these powerful tools, we are continuing to explore the vast potential of LLMs in advancing scientific research while mitigating the risks associated with their misuse. A brief dual-use study of Coscientist is provided in Supplementary Information section ‘ Safety implications: Dual-use study’ .

Technology use disclosure

The writing of the preprint version of this manuscript was assisted by ChatGPT (specifically, GPT-4 being used for grammar and typos). All authors have read, corrected and verified all information presented in this manuscript and Supplementary Information.

Data availability

Examples of the experiments discussed in the text are provided in the Supplementary Information . Because of safety concerns, data, code and prompts will be only fully released after the development of US regulations in the field of artificial intelligence and its scientific applications. Nevertheless, the outcomes of this work can be reproduced using actively developed frameworks for autonomous agent development. The reviewers had access to the web application and were able to verify any statements related to this work. Moreover, we provide a simpler implementation of the described approach, which, although it may not produce the same results, allows for deeper understanding of the strategies used in this work.

Code availability

Simpler implementation as well as generated outputs used for quantitative analysis are provided at https://github.com/gomesgroup/coscientist .

Brown, T. et al. in Advances in Neural Information Processing Systems Vol. 33 (eds Larochelle, H. et al.) 1877–1901 (Curran Associates, 2020).

Thoppilan, R. et al. LaMDA: language models for dialog applications. Preprint at https://arxiv.org/abs/2201.08239 (2022).

Touvron, H. et al. LLaMA: open and efficient foundation language models. Preprint at https://arxiv.org/abs/2302.13971 (2023).

Hoffmann, J. et al. Training compute-optimal large language models. In Advances in Neural Information Processing Systems 30016–30030 (NeurIPS, 2022).

Chowdhery, A. et al. PaLM: scaling language modeling with pathways. J. Mach. Learn. Res. 24 , 1–113 (2022).

Lin, Z. et al. Evolutionary-scale prediction of atomic-level protein structure with a language model. Science 379 , 1123–1130 (2023).

Article   ADS   MathSciNet   CAS   PubMed   Google Scholar  

Luo, R. et al. BioGPT: generative pre-trained transformer for biomedical text generation and mining. Brief Bioinform. 23 , bbac409 (2022).

Article   PubMed   Google Scholar  

Irwin, R., Dimitriadis, S., He, J. & Bjerrum, E. J. Chemformer: a pre-trained transformer for computational chemistry. Mach. Learn. Sci. Technol. 3 , 015022 (2022).

Article   ADS   Google Scholar  

Kim, H., Na, J. & Lee, W. B. Generative chemical transformer: neural machine learning of molecular geometric structures from chemical language via attention. J. Chem. Inf. Model. 61 , 5804–5814 (2021).

Article   CAS   PubMed   Google Scholar  

Jablonka, K. M., Schwaller, P., Ortega-Guerrero, A. & Smit, B. Leveraging large language models for predictive chemistry. Preprint at https://chemrxiv.org/engage/chemrxiv/article-details/652e50b98bab5d2055852dde (2023).

Xu, F. F., Alon, U., Neubig, G. & Hellendoorn, V. J. A systematic evaluation of large language models of code. In Proc. 6th ACM SIGPLAN International Symposium on Machine Programming 1–10 (ACM, 2022).

Nijkamp, E. et al. CodeGen: an open large language model for code with multi-turn program synthesis. In Proc. 11th International Conference on Learning Representations (ICLR, 2022).

Kaplan, J. et al. Scaling laws for neural language models. Preprint at https://arxiv.org/abs/2001.08361 (2020).

OpenAI. GPT-4 Technical Report (OpenAI, 2023).

Ziegler, D. M. et al. Fine-tuning language models from human preferences. Preprint at https://arxiv.org/abs/1909.08593 (2019).

Ouyang, L. et al. Training language models to follow instructions with human feedback. In Advances in Neural Information Processing Systems 27730–27744 (NeurIPS, 2022).

Granda, J. M., Donina, L., Dragone, V., Long, D.-L. & Cronin, L. Controlling an organic synthesis robot with machine learning to search for new reactivity. Nature 559 , 377–381 (2018).

Article   ADS   CAS   PubMed   PubMed Central   Google Scholar  

Caramelli, D. et al. Discovering new chemistry with an autonomous robotic platform driven by a reactivity-seeking neural network. ACS Cent. Sci. 7 , 1821–1830 (2021).

Article   CAS   PubMed   PubMed Central   Google Scholar  

Angello, N. H. et al. Closed-loop optimization of general reaction conditions for heteroaryl Suzuki–Miyaura coupling. Science 378 , 399–405 (2022).

Adamo, A. et al. On-demand continuous-flow production of pharmaceuticals in a compact, reconfigurable system. Science 352 , 61–67 (2016).

Article   ADS   CAS   PubMed   Google Scholar  

Coley, C. W. et al. A robotic platform for flow synthesis of organic compounds informed by AI planning. Science 365 , eaax1566 (2019).

Burger, B. et al. A mobile robotic chemist. Nature 583 , 237–241 (2020).

Auto-GPT: the heart of the open-source agent ecosystem. GitHub https://github.com/Significant-Gravitas/AutoGPT (2023).

BabyAGI. GitHub https://github.com/yoheinakajima/babyagi (2023).

Chase, H. LangChain. GitHub https://github.com/langchain-ai/langchain (2023).

Bran, A. M., Cox, S., White, A. D. & Schwaller, P. ChemCrow: augmenting large-language models with chemistry tools. Preprint at https://arxiv.org/abs/2304.05376 (2023).

Liu, P. et al. Pre-train, prompt, and predict: a systematic survey of prompting methods in natural language processing. ACM Comput. Surv. 55 , 195 (2021).

Bai, Y. et al. Constitutional AI: harmlessness from AI feedback. Preprint at https://arxiv.org/abs/2212.08073 (2022).

Falcon LLM. TII https://falconllm.tii.ae (2023).

Open LLM Leaderboard. Hugging Face https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard (2023).

Ji, Z. et al. Survey of hallucination in natural language generation. ACM Comput. Surv. 55 , 248 (2023).

Article   Google Scholar  

Reaxys https://www.reaxys.com (2023).

SciFinder https://scifinder.cas.org (2023).

Yao, S. et al. ReAct: synergizing reasoning and acting in language models. In Proc.11th International Conference on Learning Representations (ICLR, 2022).

Wei, J. et al. Chain-of-thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems 24824–24837 (NeurIPS, 2022).

Long, J. Large language model guided tree-of-thought. Preprint at https://arxiv.org/abs/2305.08291 (2023).

Opentrons Python Protocol API. Opentrons https://docs.opentrons.com/v2/ (2023).

Tu, Z. et al. Approximate nearest neighbor search and lightweight dense vector reranking in multi-stage retrieval architectures. In Proc. 2020 ACM SIGIR on International Conference on Theory of Information Retrieval 97–100 (ACM, 2020).

Lin, J. et al. Pyserini: a python toolkit for reproducible information retrieval research with sparse and dense representations. In Proc. 44th International ACM SIGIR Conference on Research and Development in Information Retrieval 2356–2362 (ACM, 2021).

Qadrud-Din, J. et al. Transformer based language models for similar text retrieval and ranking. Preprint at https://arxiv.org/abs/2005.04588 (2020).

Paper QA. GitHub https://github.com/whitead/paper-qa (2023).

Robertson, S. & Zaragoza, H. The probabilistic relevance framework: BM25 and beyond. Found. Trends Inf. Retrieval 3 , 333–389 (2009).

Data Mining. Mining of Massive Datasets (Cambridge Univ., 2011).

Johnson, J., Douze, M. & Jegou, H. Billion-scale similarity search with GPUs. IEEE Trans. Big Data 7 , 535–547 (2021).

Vechtomova, O. & Wang, Y. A study of the effect of term proximity on query expansion. J. Inf. Sci. 32 , 324–333 (2006).

Running experiments. Emerald Cloud Lab https://www.emeraldcloudlab.com/guides/runningexperiments (2023).

Sanchez-Garcia, R. et al. CoPriNet: graph neural networks provide accurate and rapid compound price prediction for molecule prioritisation. Digital Discov. 2 , 103–111 (2023).

Bubeck, S. et al. Sparks of artificial general intelligence: early experiments with GPT-4. Preprint at https://arxiv.org/abs/2303.12712 (2023).

Ramos, M. C., Michtavy, S. S., Porosoff, M. D. & White, A. D. Bayesian optimization of catalysts with in-context learning. Preprint at https://arxiv.org/abs/2304.05341 (2023).

Perera, D. et al. A platform for automated nanomole-scale reaction screening and micromole-scale synthesis in flow. Science 359 , 429–434 (2018).

Ahneman, D. T., Estrada, J. G., Lin, S., Dreher, S. D. & Doyle, A. G. Predicting reaction performance in C–N cross-coupling using machine learning. Science 360 , 186–190 (2018).

Hickman, R. et al. Atlas: a brain for self-driving laboratories. Preprint at https://chemrxiv.org/engage/chemrxiv/article-details/64f6560579853bbd781bcef6 (2023).

Download references


We thank the following Carnegie Mellon University Chemistry groups for their assistance with providing the chemicals needed for the Coscientist’s experiments: Sydlik, Garcia Borsch, Matyjaszewski and Ly. We give special thanks to the Noonan group (K. Noonan and D. Sharma) for providing access to chemicals and gas chromatography–mass spectrometry analysis. We also thank the team at Emerald Cloud Lab (with special attention to Y. Benslimane, H. Gronlund, B. Smith and B. Frezza) for assisting us with parsing their documentation and executing experiments. G.G. is grateful to the Carnegie Mellon University Cloud Lab Initiative led by the Mellon College of Science for its vision of the future of physical sciences. G.G. thanks Carnegie Mellon University; the Mellon College of Sciences and its Department of Chemistry; and the College of Engineering and its Department of Chemical Engineering for the start-up support. D.A.B. was partially funded by the National Science Foundation Center for Chemoenzymatic Synthesis (Grant no. 2221346). R.M. was funded by the National Science Foundation Center for Computer-Assisted Synthesis (Grant no. 2202693).

Author information

Authors and affiliations.

Department of Chemical Engineering, Carnegie Mellon University, Pittsburgh, PA, USA

Daniil A. Boiko, Robert MacKnight & Gabe Gomes

Emerald Cloud Lab, South San Francisco, CA, USA

Department of Chemistry, Carnegie Mellon University, Pittsburgh, PA, USA

Wilton E. Scott Institute for Energy Innovation, Carnegie Mellon University, Pittsburgh, PA, USA

You can also search for this author in PubMed   Google Scholar


D.A.B. designed the computational pipeline and developed the ‘Planner’, ‘Web searcher’ and ‘Code execution’ modules. R.M. assisted in designing the computational pipeline and developed the ‘Docs searcher’ module. B.K. analysed the behaviours of the Docs searcher module to enable Coscientist to produce experiment code in Emerald Cloud Lab’s Symbolic Lab Language. D.A.B. assisted and oversaw Coscientist’s chemistry experiments. D.A.B., R.M. and G.G. designed and performed initial computational safety studies. D.A.B. designed and graded Coscientist’s synthesis capabilities study. D.A.B. co-designed with G.G. and performed the optimization experiments. R.M. performed the large compound library experiment and Bayesian optimization baseline runs. G.G. designed the concepts, performed preliminary studies and supervised the project. D.A.B., R.M. and G.G. wrote this manuscript.

Corresponding author

Correspondence to Gabe Gomes .

Ethics declarations

Competing interests.

G.G. is part of the AI Scientific Advisory Board of Emerald Cloud Lab. Experiments and conclusions in this manuscript were made before G.G.’s appointment to this role. B.K. is an employee of Emerald Cloud Lab. D.A.B. and G.G. are co-founders of aithera.ai, a company focusing on responsible use of artificial intelligence for research.

Peer review

Peer review information.

Nature thanks Sebastian Farquhar, Tiago Rodrigues and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended data fig. 1 using uv-vis and liquid handler to solve food colouring identification problem..

Guiding prompt in the third message is shown in bold. In the first message the user prompt is provided, then code for sample preparation is generated, resulting data is provided as NumPy array, which is then analysed to give the final answer.

Extended Data Fig. 2 Code, generated by Coscientist.

The generated code can be split into the following steps: defining metadata for the method, loading labware modules, setting up the liquid handler, performing required reagent transfers, setting up the heater-shaker module, running the reaction, and turning the module off.

Extended Data Fig. 3 Additional results on comparison with Bayesian optimization.

a , GPT-4 models compared with Bayesian optimization performed starting with different number of initial samples. b , Compound-by-compound comparison of differences between advantages.

Supplementary information

Supplementary information.

Supplementary Text and Figs. 1–3.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Boiko, D.A., MacKnight, R., Kline, B. et al. Autonomous chemical research with large language models. Nature 624 , 570–578 (2023). https://doi.org/10.1038/s41586-023-06792-0

Download citation

Received : 20 April 2023

Accepted : 27 October 2023

Published : 20 December 2023

Issue Date : 21 December 2023

DOI : https://doi.org/10.1038/s41586-023-06792-0

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

This article is cited by

This gpt-powered robot chemist designs reactions and makes drugs — on its own.

  • Katharine Sanderson

Nature (2023)

Large language models direct automated chemistry laboratory

  • Ana Laura Dias
  • Tiago Rodrigues

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

  • Explore articles by subject
  • Guide to authors
  • Editorial policies

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

writing for the journal of orthopaedic research

  • Open Advanced Search
  • Enterprise Plans

Get 20M+ Full-Text Papers For Less Than $1.50/day.  Start a 14-Day Trial for You or Your Team.

Learn more →.

Sorry, your browser isn't supported.

Upgrading to a modern browser will give you the best experience with DeepDyve.

  • Google Chrome

Writing for the Journal of Orthopaedic Research

  • References BETA
  • Recommended
  • Add to Folder
  • Times Cited: Web of Science

 *Editor, Journal of Orthopaedic Research; ?Hospital for Special Surgery, New York, New York; $Department of Orthopaedic Surgery, University of Iowa Hospitals and Clinics, Iowa City, Iowa; and §Vice Provost for Research, Oregon State University, Cowallis, Oregon, U S A . Good scientific writing requires good science and good writing. Unfortunately, the last time most of us were asked to think about the mechanics of writing was in grade school. As a result, many of us have forgotten the rules of grammar, the weakness of the passive voice, and the need for topic sentences and transitional phrases in the construction of a paragraph. In addition, few of us have been taught to write a scientific manuscript. Instead, we learn by emulating available (and sometimes imperfect) literature and by the slow and often painful process of writing and publishing our work. Furthermore, of the many texts and articles about scientific writing, few deal in practical terms with the form and content of biomedical research papers. Thus, when planning to publish our research results, we can be faced with a series of questions. What should be included in the Introduction? How much literature should be reviewed? How many reference citations are

Have problems reading an article? Let us know here.

Thanks for helping us catch any problems with articles on DeepDyve. We'll do our best to fix them.

How was the reading experience on this article?

Check all that apply - Please note that only the first page is available if you have not selected a reading option after clicking "Read Article".

Include any more information that will help us locate the issue and fix it faster for you.

Thank you for submitting a report!

Submitting a report will send us an email through our customer support system.

References (5)

Buckwalter Buckwalter, Wright Wright, Frank Frank, Martin Martin, Sandell Sandell, Trippel Trippel (1997)

J Orthop Res , 15

J. Buckwalter, T. Wright, C. Frank, R. Martin, L. Sandell, S. Trippel (1997)

Journal of Orthopaedic Research , 15

M. Zeiger (1991)

  • Journal of Orthopaedic Research® /
  • Volume 17 Issue 4
  • Subject Areas /
  • Orthopedics and Sports Medicine

*Editor, Journal of Orthopaedic Research; ?Hospital for Special Surgery, New York, New York; $Department of Orthopaedic Surgery, University of Iowa Hospitals and Clinics, Iowa City, Iowa; and §Vice Provost for Research, Oregon State University, Cowallis, Oregon, U S A . Good scientific writing requires good science and good writing. Unfortunately, the last time most of us were asked to think about the mechanics of writing was in grade school. As a result, many of us have forgotten the rules of grammar, the weakness of the passive voice, and the need for topic sentences and transitional phrases in the construction of a paragraph. In addition, few of us have been taught to write a scientific manuscript. Instead, we learn by emulating available (and sometimes imperfect) literature and by the slow and often painful process of writing and publishing our work. Furthermore, of the many texts and articles about scientific writing, few deal in practical terms with the form and content of biomedical research papers. Thus, when planning to publish our research results, we can be faced with a series of questions. What should be included in the Introduction? How much literature should be reviewed? How many reference citations are

Journal of Orthopaedic Research – Wiley

Published: Jul 1, 1999

Recommended Articles

There are no references for this article., share the full text of this article with up to 5 colleagues for free, sign up for your 14-day free trial now.

Read and print from thousands of top scholarly journals.

Continue with Facebook

Log in with Microsoft

Already have an account? Log in

Save Article to Bookmarks

Bookmark this article. You can see your Bookmarks on your DeepDyve Library .

To save an article, log in first, or sign up for a DeepDyve account if you don’t already have one.

Sign Up Log In

Subscribe to Journal Email Alerts

To subscribe to email alerts, please log in first, or sign up for a DeepDyve account if you don’t already have one.

Copy and paste the desired citation format or use the link below to download a file formatted for EndNote

Reference Managers

Follow a Journal

To get new article updates from a journal on your personalized homepage, please log in first, or sign up for a DeepDyve account if you don’t already have one.

writing for the journal of orthopaedic research

Access the full text.

Sign up today, get DeepDyve free for 14 days.

Our policy towards the use of cookies

All DeepDyve websites use cookies to improve your online experience. They were placed on your computer when you launched this website. You can change your cookie settings through your browser.

cookie policy


  1. 2013 Edition of Journal of Orthopaedic Surgery.

    writing for the journal of orthopaedic research

  2. Journal of Trauma & Orthopaedics

    writing for the journal of orthopaedic research

  3. Subscribe to Journal of Orthopaedics

    writing for the journal of orthopaedic research

  4. Journal of Orthopaedic Research : Vol 32 , No 1

    writing for the journal of orthopaedic research

  5. Journal of Orthopaedics, Trauma and Rehabilitation: SAGE Journals

    writing for the journal of orthopaedic research

  6. Orthopedic Trauma Journals

    writing for the journal of orthopaedic research


  1. Writing

  2. TR Webinar -Writing High-Quality Manuscripts and Publishing Your Research

  3. Gartland classification for Pediatric Supracondylar fractures

  4. Carpal Coalation

  5. Orthopaedic Entrepreneurship 1: Getting to the Finish Line

  6. Case report of hip PVNS


  1. Journal of Orthopaedic Research

    How to Submit Manuscript Preparation Author Formatting Titles and Abstracts English Language Author Information Overall Formatting Details of Style Clinical Trials I nstitutional Review Board Manuscript Text Title Page Author Contribution Statement Abstract and Key Words Methods Acknowledgements References Figure Legends Tables Figures

  2. PDF Writing for the Journal of Orthopaedic Research

    U.S.A. E-mail: [email protected] repeat the same writing guidelines again and again. One of the most distinctive features of the Journal of Orthopaedic Research is its multidisciplinary read- ership. Readers vary considerably in their level of knowledge concerning the structure and function of the musculoskeletal system.

  3. Writing for the Journal of Orthopaedic Research

    Writing for the Journal of Orthopaedic Research Writing for the Journal of Orthopaedic Research J Orthop Res. 1999 Jul;17(4):459-66.doi: 10.1002/jor.1100170402. Authors T M Wright 1 , J A Buckwalter, W C Hayes Affiliation

  4. Writing for the Journal of Orthopaedic Research

    Preview Available Scholarly Journal Writing for the Journal of Orthopaedic Research Wright, Timothy M; Buckwalter, Joseph A; Hayes, Wilson C. Journal of Orthopaedic Research; Hoboken Vol. 17, Iss. 4, (Jul 1999): 459-66. Copy Link CiteAll Options You might have access to the full article...

  5. Writing for the Journal of Orthopaedic Research

    The Journal of Orthopaedic Research, a publication of the Orthopaedic Research Society (ORS), is the forum for the rapid publication of high quality reports of new information on the full spectrum of orthopaedic research, including life sciences, engineering, translational, and clinical studies.

  6. The ORS Journals: The Journal of Orthopaedic Research (JOR)

    The Journal of Orthopaedic Research (JOR) JOR publishes reports on the full spectrum of orthopaedic research, including: life sciences, engineering, translational, and clinical studies. Accepted Articles Early View Current Issue All Issues JOR Special Issues Virtual Issues Perspective Articles Journal of Orthopaedic Research

  7. Overview

    The Journal of Orthopaedic Research, a publication of the Orthopaedic Research Society (ORS), is the forum for the rapid publication of high quality reports of new information on the full spectrum of orthopaedic research, including life sciences, engineering, translational, and clinical studies.

  8. Writing for Clinical Orthopaedics and Related Research

    Styles of writing are as numerous as authors, although most journals publish guidelines for formatting a manuscript, and many have more or less established writing styles (eg, the American Medical Association Manual of Style) [ 3 ]. CORR uses the AMA style as a general guideline.

  9. Setting standards for medical writing in orthopaedics

    Read. The most prolific scientific writers are often the most avid readers. In addition to improving writing skills, reading enables you to identify gaps in knowledge, to be familiar with the literature in the field and to highlight hot topics that may become sources of ideas for future research [].Controversial topics or ideas may be recognized in certain manuscripts, such as letters to the ...

  10. Writing for the Journal of Orthopaedic Research

    T. Wright, J. Buckwalter, W. Hayes. Published in Journal of Orthopaedic… 1 July 1999. Education. TLDR. Of the many texts and articles about scientific writing, few deal in practical terms with the form and content of biomedical research papers, so when planning to publish research results, authors can be faced with a series of questions. Expand.

  11. How to Write a Case Report for Journal of Orthopaedic Case Reports

    More than two years have passed since Journal of Orthopaedic Case Reports (JOCR) has started publication. Over this period of time a huge number of reviewers and Editorial Board Members have contributed to review the articles and send comments on improvement of various aspect of case reports.

  12. How to write an original article for the Journal of Orthopaedics and

    An original article (OA) is the publication of a study. Thus, it might be alternatively named "study report", by analogy with "case report". The Journal of Orthopaedics and Traumatology welcomes OAs based on clinical or preclinical studies relevant to the musculoskeletal system. Although medical research methodology is not the object of ...

  13. Writing for the Journal of Orthopaedic Research.

    Journal of Orthopaedic Research : Official Publication of the Orthopaedic Research Society, 01 Jul 1999, 17(4): 459-466 DOI: 10.1002/jor.1100170402 PMID: 10459750 Share this article Share with email Share with twitter Share with linkedin Share with facebook

  14. Journal of Orthopaedic Research

    Browse open Calls for Papers beta. Read the latest articles of Journal of Orthopaedic Research at ScienceDirect.com, Elsevier's leading platform of peer-reviewed scholarly literature.

  15. Guide for authors

    INTRODUCTION • Article categories • Address all correspondence to • Submission checklist BEFORE YOU BEGIN • Ethics in publishing • Studies in humans and animals • Declaration of interest • Declaration of generative AI in scientific writing • Submission declaration • Use of inclusive language • Reporting sex- and gender-based analyses

  16. Journal of orthopaedic research : official publication of the

    Journal of orthopaedic research : official publication of the Orthopaedic Research Society Journal Overview publication venue for . A multimodal assessment of cementless tibial baseplate fixation using radiography, radiostereometric analysis, and magnetic resonance imaging. 2023 Bone-ACL-bone allograft for anterior cruciate ligament reconstruction: Short-term evaluation in a rabbit model with ...

  17. DOCX Guidelines for Authors

    ® brings readers the latest clinical and basic research and informed opinions that shape today's orthopaedic practice, thereby providing an opportunity to practice evidence-based medicine. With contributions from leading clinicians and researchers around the world, we aim to be the world's general-interest orthopaedic journal.

  18. Journal of Orthopaedic Research

    December 1, 2022 The Journal of Orthopaedic Research ® (JOR) is pleased to announce a Special Issue on Tendon Research: Guiding the Future with guest editors, Nathaniel Dyment, PhD, Andrew Kuntz, MD, and Louis Soslowsky, PhD from the University of Pennsylvania. We welcome Original Papers in any topic related to tendon.

  19. Manuscript Submission Guidelines: Journal of Orthopaedics, Trauma and

    Journal of Orthopaedics, Trauma and Rehabilitation may accept submissions of papers that have been posted on pre-print servers; please alert the Editorial Office when submitting (contact details are at the end of these guidelines) and include the DOI for the preprint in the designated field in the manuscript submission system.

  20. Journal of Orthopaedic Surgery and Research

    Journal of Orthopaedic Surgery and Research is an open access, peer-reviewed online journal that encompasses all aspects of clinical and basic research studies related to musculoskeletal issues.. Journal of Orthopaedic Surgery and Research provides the platform for exchange of new clinical and scientific information in the most precise and expeditious way to achieve timely dissemination of ...

  21. Writing For Journal of Orthopaedic Research PDF

    WRITING FOR T H E JOURNAL OF ORTHOPAEDIC RESEARCH 461. tion (for example, use of ACL for anterior cruciate Just as scntences should be short and not over-ligament); however, without appropriate definitions, loaded with information, paragraphs should be as they can confuse the reader.

  22. Claudine Gay and Why Academic Honesty Matters

    Cambridge, Mass. Claudine Gay, the president of my university, is under attack for academic dishonesty. She is charged with several instances of plagiarism, in her dissertation and other published ...

  23. Autonomous chemical research with large language models

    Coscientist is an artificial intelligence system driven by GPT-4 that autonomously designs, plans and performs experiments by incorporating large language models empowered by tools such as ...

  24. Writing for the Journal of Orthopaedic Research

    Writing for the Journal of Orthopaedic Research Wright, Timothy M.; Buckwalter, Joseph A.; Hayes, Wilson C. Journal of Orthopaedic Research , Volume 17 (4) - Jul 1, 1999 Read Article Download PDF Share Full Text for Free 8 pages Article Details Recommended References Bookmark Add to Folder Cite Social Times Cited: Web of Science Journals /

  25. The ORS Journals: The Journal of Orthopaedic Research

    The ORS produces two internationally renowned, peer reviewed musculoskeletal journals, the Journal of Orthopaedic Research (JOR) and the Journal of Orthopaedic Research - Spine (JORS). These journals offers a comprehensive and multidisciplinary approach to orthopaedic research, ensuring that you stay informed, connected, and contribute to the advancement of this crucial field. JOR and JORS ...