Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 73 of the 81 references it cites were found in a registry or an index; 27 of the 67 sources its claims cite were checked for support.

A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial

Open access

Abstract

Phase 2a multicenter, double-blind, randomized, placebo-controlled trial of rentosertib (formerly ISM001-055), a first-in-class AI-generated small-molecule inhibitor of TNIK for idiopathic pulmonary fibrosis. Safety was similar across arms; highest-dose group showed mean FVC increase versus decline with placebo over 12 weeks.

Overview

CoChat Summary

## TL;DR

This phase 2a randomized, double-blind, placebo-controlled trial found that the AI-discovered TNIK inhibitor rentosertib was generally safe and well tolerated in idiopathic pulmonary fibrosis (IPF) over 12 weeks. The highest dose tested, 60 mg once daily, showed the most favorable lung-function signal, including a mean FVC change of +98.4 ml versus −20.3 ml with placebo.

## Key findings

  • Among 71 randomized patients, treatment-emergent adverse events were common but similar across groups: 72.2% (13/18) with 30 mg QD, 83.3% (15/18) with 30 mg BID, 83.3% (15/18) with 60 mg QD, and 70.6% (12/17) with placebo.
  • Treatment-related serious adverse events were uncommon: 0 in placebo, 1/18 (5.6%) in 30 mg QD, and 2/18 (11.1%) in each of the 30 mg BID and 60 mg QD groups.
  • After 12 weeks, mean FVC change was −20.3 ml with placebo, −27.0 ml with 30 mg QD, +19.7 ml with 30 mg BID, and +98.4 ml with 60 mg QD (95% CI 10.9 to 185.9).
  • In the 60 mg QD group without concurrent standard-of-care antifibrotic therapy, FVC improved by +187.8 ml (95% CI 68.6 to 306.9 ml); the paper also reports a significant LCQ improvement at 60 mg QD versus placebo (two-sided P = 0.0495).

## Methods

  • Phase 2a, multicenter, double-blind, randomized, placebo-controlled, multidose trial at 21 sites in China; adults with IPF were assigned 1:1:1:1 to placebo or rentosertib 30 mg QD, 30 mg BID, or 60 mg QD for 12 weeks.
  • 128 patients were screened and 71 enrolled; the intention-to-treat population included all 71 randomized patients, and all were included in safety and efficacy analyses.
  • Primary endpoint was the proportion with at least one treatment-emergent adverse event; secondary analyses included FVC and other lung-function measures, LCQ, 6MWD, pharmacokinetics, and exploratory serum proteomics, using ANCOVA for FVC, mixed models for LCQ, multiple imputation for missing week-12 values, and generalized linear/mixed models for proteomics.

## Limitations

  • The paper states that limitations include the small cohort size in each arm, the geographical and demographic homogeneity of participants (all residents of China of similar race), and the short follow-up period, which limit assessment of long-term safety and efficacy.
  • The text makes evident that 22/71 patients were missing week-12 measurements and 16/71 (22.5%) discontinued treatment before study end, which may affect precision and completeness of efficacy estimates.
  • The 12-week duration is also explicitly noted as relatively short for capturing longer-term events such as acute exacerbations of IPF.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41591-025-03743-2

Cites

73 of 73 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Trial participant randomization and follow-up scheme. Of 128 patients screened for inclusion, 71 were randomized to receive 30 mg rentosertib QD (n = 18), 30 mg rentosertib BID (n = 18), 60 mg rentosertib QD (n = 18), or placebo (n = 17) over the course of 12 weeks. Sixteen patients discontinued treatment prior to the end of treatment. Source data
Fig. 1 Trial participant randomization and follow-up scheme. Of 128 patients screened for inclusion, 71 were randomized to receive 30 mg rentosertib QD (n = 18), 30 mg rentosertib BID (n = 18), 60 mg rentosertib QD (n = 18), or placebo (n = 17) over the course of 12 weeks. Sixteen patients discontinued treatment prior to the end of treatment. Source dataCC BY-NC-ND · PubMed Central
Fig. 2 Changes in FVC ± 95% CI after 12 weeks of rentosertib treatment compared to baseline. a, The absolute change in FVC ± 95% CI. b, The absolute change in FVC ± 95% CI ANCOVA model with multiple imputation assuming missing at random (MAR).
Fig. 2 Changes in FVC ± 95% CI after 12 weeks of rentosertib treatment compared to baseline. a, The absolute change in FVC ± 95% CI. b, The absolute change in FVC ± 95% CI ANCOVA model with multiple imputation assuming missing at random (MAR).CC BY-NC-ND · PubMed Central
Fig. 3 Pharmacokinetic properties of rentosertib. a–c, Pharmacokinetic dynamics of rentosertib in patient sera collected predose and periodically throughout 24 h posttreatment at week 0 (a), at 0 h posttreatment at weeks 2, 4 and 8 (b) and 48 h after administration at week 12 (c). d, AUC0–t of rentosertib exposure; n = 22 in 30 mg QD, n = 21 in 30 mg BID, n = 20 in 60 mg QD. e, Cmax of rentosertib exposure; n = 22 in 30 mg QD, n = 21 in 30 mg BID, n = 20 in 60 mg QD. f, Ctrough of rentosertib exposure in patient sera collected pretreatment at weeks 0, 2, 4, 8 and 12. All data represent arithmetic mean ± s.d.
Fig. 3 Pharmacokinetic properties of rentosertib. a–c, Pharmacokinetic dynamics of rentosertib in patient sera collected predose and periodically throughout 24 h posttreatment at week 0 (a), at 0 h posttreatment at weeks 2, 4 and 8 (b) and 48 h after administration at week 12 (c). d, AUC0–t of rentosertib exposure; n = 22 in 30 mg QD, n = 21 in 30 mg BID, n = 20 in 60 mg QD. e, Cmax of rentosertib exposure; n = 22 in 30 mg QD, n = 21 in 30 mg BID, n = 20 in 60 mg QD. f, Ctrough of rentosertib exposure in patient sera collected pretreatment at weeks 0, 2, 4, 8 and 12. All data represent arithmetic mean ± s.d.CC BY-NC-ND · PubMed Central
Fig. 4 Post-hoc exploratory analysis of serum protein profiling. a,b, Differentially abundant proteins in serum from patients receiving rentosertib at 30 mg BID (a) and 60 mg QD (b) identified with a generalized linear model. c, Serum levels of COL1A1, FAP, FN1 and MMP10 decrease with rentosertib dose and time on treatment. n = 11 patients in placebo, n = 11 patients in 30 QD, n = 11 patients in 30 BID, n = 10 patients in 60 QD. d, The changes in serum levels of COL1A1, FAP, FN1 and MMP10 from baseline to week 12 are inversely correlated with change in FVC from baseline to week 12. n = 43 patients at each visit; P value calculated by Pearson correlation analysis. e,f, Reactome pathway enrichment for differentially abundant proteins in serum from patients receiving rentosertib at 30 mg BID (e) and 60 mg QD (f).
Fig. 4 Post-hoc exploratory analysis of serum protein profiling. a,b, Differentially abundant proteins in serum from patients receiving rentosertib at 30 mg BID (a) and 60 mg QD (b) identified with a generalized linear model. c, Serum levels of COL1A1, FAP, FN1 and MMP10 decrease with rentosertib dose and time on treatment. n = 11 patients in placebo, n = 11 patients in 30 QD, n = 11 patients in 30 BID, n = 10 patients in 60 QD. d, The changes in serum levels of COL1A1, FAP, FN1 and MMP10 from baseline to week 12 are inversely correlated with change in FVC from baseline to week 12. n = 43 patients at each visit; P value calculated by Pearson correlation analysis. e,f, Reactome pathway enrichment for differentially abundant proteins in serum from patients receiving rentosertib at 30 mg BID (e) and 60 mg QD (f).CC BY-NC-ND · PubMed Central
Extended Data Fig. 1 Key trial eligibility criteria, randomization scheme, and schedule of key study activities. Eligible patients were randomized to receive 30 mg rentosertib QD, 30 mg rentosertib BID, 60 mg rentosertib QD, or placebo administered over 12 weeks with periodic assessment and plasma sampling. IPF, idiopathic pulmonary fibrosis; FVC, forced vital capacity; FEV1, forced expiratory volume in one second; DLCO, diffusion capacity of the lung for carbon monoxide; SOC, standard-of-care; QD, once-daily; BID, twice-daily; HRCT, high-resolution computed tomography; 6MWD, six-minute walk distance; LCQ, Leicester cough questionnaire; PK, pharmacokinetics; EOT, end-of-trial; EOS, end-of-study.
Extended Data Fig. 1 Key trial eligibility criteria, randomization scheme, and schedule of key study activities. Eligible patients were randomized to receive 30 mg rentosertib QD, 30 mg rentosertib BID, 60 mg rentosertib QD, or placebo administered over 12 weeks with periodic assessment and plasma sampling. IPF, idiopathic pulmonary fibrosis; FVC, forced vital capacity; FEV1, forced expiratory volume in one second; DLCO, diffusion capacity of the lung for carbon monoxide; SOC, standard-of-care; QD, once-daily; BID, twice-daily; HRCT, high-resolution computed tomography; 6MWD, six-minute walk distance; LCQ, Leicester cough questionnaire; PK, pharmacokinetics; EOT, end-of-trial; EOS, end-of-study.CC BY-NC-ND · PubMed Central
Extended Data Fig. 2 Changes in FVC after 12 weeks of treatment with rentosertib. a-b) Changes in forced vital capacity (FVC) ± 95% CI after 12 weeks of treatment compared to baseline excluding n = 1 patient from the placebo group and n = 1 patient from the rentosertib 30 mg QD group who exhibited >600 mL difference between screening and baseline FVC measurements, making uncertain the baseline FVC values in those patients. Absolute change in FVC ± 95% CI (a) and absolute change in FVC ± 95% CI ANCOVA Model with Multiple Imputation assuming missing at random (MAR) (b). (c) Absolute change in FVC (% Predicted) ± SE, ANCOVA Model with Multiple Imputation assuming MAR. (d) Percentage change in FVC ± SE, ANCOVA Model with Multiple Imputation assuming MAR. (e) Change in percent predicted FVC ± SD.
Extended Data Fig. 2 Changes in FVC after 12 weeks of treatment with rentosertib. a-b) Changes in forced vital capacity (FVC) ± 95% CI after 12 weeks of treatment compared to baseline excluding n = 1 patient from the placebo group and n = 1 patient from the rentosertib 30 mg QD group who exhibited >600 mL difference between screening and baseline FVC measurements, making uncertain the baseline FVC values in those patients. Absolute change in FVC ± 95% CI (a) and absolute change in FVC ± 95% CI ANCOVA Model with Multiple Imputation assuming missing at random (MAR) (b). (c) Absolute change in FVC (% Predicted) ± SE, ANCOVA Model with Multiple Imputation assuming MAR. (d) Percentage change in FVC ± SE, ANCOVA Model with Multiple Imputation assuming MAR. (e) Change in percent predicted FVC ± SD.CC BY-NC-ND · PubMed Central
Extended Data Fig. 3 Changes in FVC after 12 weeks of treatment with rentosertib stratified by concurrent use of SOC antifibrotic therapy. (a) Absolute change in FVC ± 95% CI in patients not concurrently taking SOC antifibrotic therapy (left) or in patients concurrently taking antifibrotic therapy (right). (b) Absolute change in FVC ± 95% CI by ANCOVA Model with Multiple Imputation assuming missing at random (MAR) in patients not concurrently taking SOC antifibrotic therapy (left) or in patients concurrently taking antifibrotic therapy (right).
Extended Data Fig. 3 Changes in FVC after 12 weeks of treatment with rentosertib stratified by concurrent use of SOC antifibrotic therapy. (a) Absolute change in FVC ± 95% CI in patients not concurrently taking SOC antifibrotic therapy (left) or in patients concurrently taking antifibrotic therapy (right). (b) Absolute change in FVC ± 95% CI by ANCOVA Model with Multiple Imputation assuming missing at random (MAR) in patients not concurrently taking SOC antifibrotic therapy (left) or in patients concurrently taking antifibrotic therapy (right).CC BY-NC-ND · PubMed Central
Extended Data Fig. 4 Changes in additional lung function metrics after 12 weeks of treatment with rentosertib. (a) Absolute change in DLCO ± 95% CI. (b) Absolute Change in DLCO (% Predicted) ± SE, ANCOVA Model with Multiple Imputation assuming missing at random (MAR). (c) Absolute change in percent predicted HGB-corrected DLCO (%) ± SD. (d) Absolute change in FEV1 ± SD. (e) Mean percent-change in FEV1 ± SD. (f) Absolute change in LCQ score ± SD. (g) Absolute change in LCQ score ± SE by Mixed model for repeated measures (MMRM). (h) Absolute change in 6MWD ± SD. (i) Absolute change in 6MWD ± SE, ANCOVA Model with Multiple Imputation assuming MAR.
Extended Data Fig. 4 Changes in additional lung function metrics after 12 weeks of treatment with rentosertib. (a) Absolute change in DLCO ± 95% CI. (b) Absolute Change in DLCO (% Predicted) ± SE, ANCOVA Model with Multiple Imputation assuming missing at random (MAR). (c) Absolute change in percent predicted HGB-corrected DLCO (%) ± SD. (d) Absolute change in FEV1 ± SD. (e) Mean percent-change in FEV1 ± SD. (f) Absolute change in LCQ score ± SD. (g) Absolute change in LCQ score ± SE by Mixed model for repeated measures (MMRM). (h) Absolute change in 6MWD ± SD. (i) Absolute change in 6MWD ± SE, ANCOVA Model with Multiple Imputation assuming MAR.CC BY-NC-ND · PubMed Central
Extended Data Fig. 5 Association of rentosertib treatment with PK parameters. (a) AUC0-t of rentosertib at week 0 and week 12, data are mean ± SD. n = 11 at week 0 and n = 11 at week 12 in 30QD, n = 11 at week 0 and n = 10 at week 12 in 30 BID, n = 10 at week 0 and n = 10 at week 10 in 60 QD. (b) t1/2 of rentosertib at week 0 and week 12, data are mean ± SD. n = 11 at week 0 and n = 11 at week 12 in 30QD, n = 11 at week 0 and n = 10 at week 12 in 30 BID, n = 10 at week 0 and n = 10 at week 10 in 60 QD. (c) Correlation analysis of AUC0-t and change in FVC between week 0 and week 12. (d) Correlation analysis of Ctrough measured throughout the trial and change in FVC between week 0 and week 12. P values calculated by two-sided Spearman correlation analysis.
Extended Data Fig. 5 Association of rentosertib treatment with PK parameters. (a) AUC0-t of rentosertib at week 0 and week 12, data are mean ± SD. n = 11 at week 0 and n = 11 at week 12 in 30QD, n = 11 at week 0 and n = 10 at week 12 in 30 BID, n = 10 at week 0 and n = 10 at week 10 in 60 QD. (b) t1/2 of rentosertib at week 0 and week 12, data are mean ± SD. n = 11 at week 0 and n = 11 at week 12 in 30QD, n = 11 at week 0 and n = 10 at week 12 in 30 BID, n = 10 at week 0 and n = 10 at week 10 in 60 QD. (c) Correlation analysis of AUC0-t and change in FVC between week 0 and week 12. (d) Correlation analysis of Ctrough measured throughout the trial and change in FVC between week 0 and week 12. P values calculated by two-sided Spearman correlation analysis.CC BY-NC-ND · PubMed Central
Extended Data Fig. 6 Time- and dose-associated changes in serum protein expression associated with rentosertib treatment. (a) Change in Normalized Protein eXpression (NPX) from baseline to subsequent visits (weeks 4, 8, and 12) with a simple linear regression model (delta_NPX ~ treatment_cat) fitted to assess the relationship between treatment dose and protein expression changes using Benjamini-Hochberg (BH)-adjusted P values. (b) Change in NPX between baseline and week 12 by two-sided paired BH-adjusted t-test. Red dots denote proteins with BH-adjusted P value < 0.1. Yellow boxes highlight fibrosis-related proteins.
Extended Data Fig. 6 Time- and dose-associated changes in serum protein expression associated with rentosertib treatment. (a) Change in Normalized Protein eXpression (NPX) from baseline to subsequent visits (weeks 4, 8, and 12) with a simple linear regression model (delta_NPX ~ treatment_cat) fitted to assess the relationship between treatment dose and protein expression changes using Benjamini-Hochberg (BH)-adjusted P values. (b) Change in NPX between baseline and week 12 by two-sided paired BH-adjusted t-test. Red dots denote proteins with BH-adjusted P value < 0.1. Yellow boxes highlight fibrosis-related proteins.CC BY-NC-ND · PubMed Central
Extended Data Fig. 7 Serum proteins with changes in expression associated with changes in FVC with rentosertib treatment. (a) PandaOmics analysis of published RNA-seq gene expression datasets profiling patients with IPF and healthy individuals of four top proteomics hit genes in our trial cohort. (b) Correlation analysis of protein abundance of genes previously associated with lung function and IPF TFS with change in FVC from week 0 to week 12, treatment regimen, and time on treatment. n = 43 patient at each visit, and the P value was calculated by two-sided Pearson correlation analysis; n = 11 patients in Placebo, n = 11 patients in 30 QD; n = 11 patients in 30 BID, n = 10 patients in 60 QD. P values calculated by paired t-test.
Extended Data Fig. 7 Serum proteins with changes in expression associated with changes in FVC with rentosertib treatment. (a) PandaOmics analysis of published RNA-seq gene expression datasets profiling patients with IPF and healthy individuals of four top proteomics hit genes in our trial cohort. (b) Correlation analysis of protein abundance of genes previously associated with lung function and IPF TFS with change in FVC from week 0 to week 12, treatment regimen, and time on treatment. n = 43 patient at each visit, and the P value was calculated by two-sided Pearson correlation analysis; n = 11 patients in Placebo, n = 11 patients in 30 QD; n = 11 patients in 30 BID, n = 10 patients in 60 QD. P values calculated by paired t-test.CC BY-NC-ND · PubMed Central
Extended Data Fig. 8 Correlation analysis of protein abundance of genes previously associated with aging-related dysfunction. Correlation analysis of protein abundance of genes (IL10, CD5, COL6A3) previously associated with aging-related dysfunction, such as immunity and ECM remodeling, with change in FVC from week 0 to week 12 and time on treatment. Left panel: n = 43 patient at each visit; P values calculated by two-sided Pearson correlation analysis; Right panel: n = 11 patients in placebo, n = 11 patients in 30 QD; n = 11 patients in 30 BID, n = 10 patients in 60 QD, boxes and whiskers are quartiles; P values calculated by paired t-test.
Extended Data Fig. 8 Correlation analysis of protein abundance of genes previously associated with aging-related dysfunction. Correlation analysis of protein abundance of genes (IL10, CD5, COL6A3) previously associated with aging-related dysfunction, such as immunity and ECM remodeling, with change in FVC from week 0 to week 12 and time on treatment. Left panel: n = 43 patient at each visit; P values calculated by two-sided Pearson correlation analysis; Right panel: n = 11 patients in placebo, n = 11 patients in 30 QD; n = 11 patients in 30 BID, n = 10 patients in 60 QD, boxes and whiskers are quartiles; P values calculated by paired t-test.CC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

20 Valid · 2 Not valid · 13 Not checked · 70 Own results and interpretations

33 more claims

Create a free account to see all 105 claims and to verify the 12 references still open.

See all claims — free account
  • It may state two separate claims as one. 42
  • Probably not a substantial claim. 36

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Show all 81 references

Cited by

See all 180

Metadata

Authors
Xu, Z., Ren, F., Wang, P., Cao, J., Tan, C., Ma, D., Zhao, L., Dai, J., Ding, Y., Fang, H., Li, H., Liu, H., Luo, F., Meng, Y., Pan, P., Xiang, P., Xiao, Z., Rao, S., Satler, C., Liu, S., Lv, Y., Zhao, H., Chen, S., Cui, H., Korzinkin, M., Gennert, D., & Zhavoronkov, A.
Affiliations
  • Peking Union Medical College Hospital (CN)
  • Tianjin Medical University General Hospital (CN)
  • Capital Medical University (CN)
  • Qilu Hospital of Shandong University (CN)
  • Sheng Jing Hospital (CN)
  • Nanjing Drum Tower Hospital (CN)
  • Hainan General Hospital (CN)
  • Tongji University (CN)
  • First Affiliated Hospital of Zhengzhou University (CN)
  • Sichuan University (CN)
  • Nanfang Hospital (CN)
  • Central South University (CN)
  • Peking University Shougang Hospital (CN)
  • Jiangxi Provincial People's Hospital (CN)
  • Insilico Medicine (United States) (US)
Corresponding author
  • Xu, Zuojun
  • Zhavoronkov, Alex
Venue
Nature Medicine
Published
2025-06-03
Volume
31
Issue
8
Pages
2602-2610
ISSN
  • 1078-8956 (ISSN-L)
  • 1546-170X
OpenAlex
W4410974018
Cited by
180
References
81
Citations by year
2026: 142 · 2025: 38
FWCI
117.24
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Medicine (publishedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
Topics
  • Health Sciences › Medicine › Pulmonary and Respiratory Medicine › Interstitial Lung Diseases and Idiopathic Pulmonary Fibrosis
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
  • Health Sciences › Medicine › Pulmonary and Respiratory Medicine › Chronic Obstructive Pulmonary Disease (COPD) Research
Keywords
idiopathic pulmonary fibrosis, small molecule inhibitors, lung function, forced vital capacity, pharmacokinetics, safety and tolerability, acute exacerbation, Leicester Cough Questionnaire, 6-minute walk distance, hepatotoxicity, diarrhea, dose comparison
MeSH terms
Aged, Artificial Intelligence, Double-Blind Method, Female, Humans, Male, Middle Aged, Treatment Outcome, Idiopathic Pulmonary Fibrosis

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial. https://app.cochat.ai/papers/10.1038/s41591-025-03743-2
The paper
Xu, Z., Ren, F., Wang, P., Cao, J., Tan, C., Ma, D., Zhao, L., Dai, J., Ding, Y., Fang, H., Li, H., Liu, H., Luo, F., Meng, Y., Pan, P., Xiang, P., Xiao, Z., Rao, S., Satler, C., . . . Zhavoronkov, A. (2025). A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial. Nature Medicine. https://doi.org/10.1038/s41591-025-03743-2

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.