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

Name: Verified Rigor Record — A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial
Item: A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial
Rating: 100
Author: Αlpha¹

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.

Paper source

Open DOI

Αlpha¹ Rigor Review

Rigor Review Verified

Critical

High-priority

Action items to resolve · Critical / High

Confidence: highEngine: 7.5.0Source: full text

Not verified in this report

References were not verified against Crossref/OpenAlex.

This Alpha1 Rigor Review uses AI Rigor Reviewers trained on a curated corpus of high-fidelity and retracted papers, with expert supervision and curation. It can still make mistakes; verify each finding against the source before relying on it.

This phase 2a clinical trial is methodologically robust, with clear randomisation, blinding, and comprehensive reporting of demographics and data/code availability. Key weaknesses include missing specific ethics committee names and protocol numbers, lack of software version numbers, inconsistent exact p-value reporting, and several copyediting inconsistencies — notably a discrepancy in the ALT/AE counts between text and Table 2.

Three independent reviewers evaluated all eight dimensions. The statistical verification component covered only 2 tests (limited coverage); the citation component found no retracted or non-existent references; the reproducibility component confirmed all 6 links live; the integrity component flagged an internal contradiction in ALT/AE reporting. Dimensions with no applicable sub-criteria (e.g., biological variables sex justified, study design replicate distinction) were excluded from score calculation.

12 major claims checked against the paper's own evidence: all adequately supported.

partialAbstract, Results, Secondary lung function endpoints
Rentosertib shows potential efficacy as measured by FVC improvement at 60 mg QD.
FVC improved by +98.4 ml (95% CI 10.9 to 185.9) in the 60 mg QD group vs -20.3 ml placebo, but this is a secondary endpoint in a small phase 2a trial; the evidence is promising but preliminary.
Evidence: Figure 2a and results text: mean change +98.4 ml (95% CI 10.9 to 185.9).
“We observed increased forced vital capacity at the highest dosage with a mean change of +98.4 ml (95% confidence interval 10.9 to 185.9) for patients in the 60 mg rentosertib QD group”
partialAbstract, Results
Rentosertib 60 mg QD improves lung function as measured by FVC over 12 weeks.
A trend toward improvement was observed (+98.4 ml, 95% CI 10.9-185.9) vs placebo decline (-20.3 ml), but the trial was not powered for efficacy and the improvement was more pronounced in patients not on SOC antifibrotics; the claim is partially supported with appropriate caveats.
Evidence: Secondary endpoint: +98.4 ml (95% CI 10.9-185.9) for 60 mg QD vs -20.3 ml for placebo.
“We observed increased forced vital capacity at the highest dosage with a mean change of +98.4 ml (95% confidence interval 10.9 to 185.9) for patients in the 60 mg rentosertib QD group, compared with −20.3 ml (95% confidence interval −116.1 to 75.6) for the placebo group.”
supportedAbstract, Results, Primary safety endpoint
Rentosertib is safe and well tolerated in patients with IPF.
Primary safety endpoint shows similar TEAE rates across groups (72-83% vs 70.6% placebo) and low treatment-related SAEs.
Evidence: Tables 2 and primary safety endpoint text.
“The primary endpoint was the percentage of patients who have at least one treatment-emergent adverse event, which was similar across all treatment arms”
supportedResults, Biomarkers and proteomic signatures
Rentosertib treatment modulates IPF pathophysiological pathways as shown by serum proteomics.
Proteomics identified downregulation of fibrosis-associated proteins (MMP10, COL1A1, etc.) and enrichment of extracellular matrix organization pathways, consistent with the hypothesized mechanism.
Evidence: Figure 4 and associated results text.
“Downregulated proteins associated with 30 mg BID and 60 mg QD treatment include known fibrosis-associated proteins such as MMP10, PTPRZ1, COL1A1, FAP, FN1, ROBO2, ASPN and LTBP2”
supportedIntroduction, paragraph 3
This is the first AI-generated small-molecule inhibitor of TNIK to enter clinical testing.
The paper states this explicitly and references prior phase 0/1 trial results.
Evidence: Introduction and earlier publications (NCT05154240).
“the first report of a targeted TNIK inhibitor entering clinical testing”
supportedIntroduction, paragraph 3
AI-enabled discovery accelerated preclinical and clinical timelines to 18 months and under 30 months respectively.
The paper states these timelines without providing further evidence within this manuscript, but earlier work is referenced.
Evidence: Introduction paragraph 3.
“Our generative AI-powered approach streamlined preclinical candidate nomination to a mere 18 months and completion of phase 0/1 clinical testing to under 30 months from the initiation of target discovery”
supportedDiscussion, paragraph 3
Treatment with 60 mg rentosertib QD over 12 weeks was associated with a trend toward an increase in FVC in patients with IPF.
The results show a mean change of +98.4 ml (95% CI 10.9 to 185.9) for the 60 mg QD group compared to -20.3 ml for placebo, which is presented as a positive trend and meets the MCID for FVC in IPF.
Evidence: Secondary lung function endpoints section, Abstract
“Treatment with 60 mg rentosertib QD over 12 weeks was associated with a trend toward an increase in FVC in patients with IPF.”
supportedDiscussion, paragraph 3
The 60 mg rentosertib QD dose exhibited the greatest mean improvement in lung function, as measured by FVC.
The reported mean change in FVC for the 60 mg QD group (+98.4 ml) is higher than for other treatment groups and placebo, supporting this claim.
Evidence: Abstract, Secondary lung function endpoints section
“The 60 mg rentosertib QD dose exhibited the greatest mean improvement in lung function, as measured by FVC, whereas patients receiving placebo experienced an average decline in FVC.”
supportedDiscussion, paragraph 6
Inhibition of TNIK modulates IPF pathophysiological pathways.
Proteomic profiling showed downregulation of fibrosis-associated proteins and enrichment of extracellular matrix organization pathways, providing evidence for the proposed mechanism of action.
Evidence: Biomarkers and proteomic signatures of response in blood section, Figure 4
“Downregulation of this IPF-associated protein profile supports the hypothesis that inhibition of TNIK modulates IPF pathophysiological pathways and points to a potential serum protein signature as a biomarker for response to rentosertib treatment.”
supportedDiscussion, paragraph 7
Using AI in both target identification and in drug design may enhance the efficiency of the drug development process.
The paper describes the AI-driven discovery of TNIK and rentosertib, and the accelerated preclinical and phase 0/1 clinical testing timelines, which supports the claim of enhanced efficiency.
Evidence: Main, paragraph 3, Discussion, paragraph 7
“Despite the short duration and the number of withdrawals from the trial across all arms ( n = 16/71 (22.5%)), the results are encouraging for additional study of this drug candidate and target, suggesting that rentosertib is generally safe and potentially effective and, more broadly, that using AI in both target identification and in drug design may enhance the efficiency of the drug development process.”
supportedAbstract, Results
Rentosertib is safe and well tolerated in patients with IPF over 12 weeks.
The primary safety endpoint (TEAE rates) was similar across treatment and placebo groups; SAEs were low and comparable. The evidence supports the claim.
Evidence: Primary safety endpoint: TEAE rates 72.2% to 83.3% vs placebo 70.6%; SAEs low (5.6-11.1% vs 0%).
“The rates of treatment-emergent AEs (TEAEs) were similar across all treatment groups”
supportedIntroduction
TNIK is a first-in-class target in idiopathic pulmonary fibrosis discovered using generative AI.
The paper cites prior work from the same group (phase 0/1 trial) establishing TNIK as a novel target for IPF discovered via their AI platform; this is a factual claim about their prior discovery process.
Evidence: Reference to prior publication: 'the first report of a targeted TNIK inhibitor entering clinical testing' and 'first reported instance of AI platform-enabled discovery of both a disease-associated target and a compound.'
“our group used generative AI-driven discovery tools ... to identify Traf2- and Nck-interacting kinase (TNIK) de novo as a critical regulator of idiopathic pulmonary fibrosis (IPF) pathology”

Found 2 reported tests, but none could be recomputed (missing degrees of freedom or sample size).

3 integrity concerns flagged (0 high).

Rigor

Eight Dimensions8 pass

12 major claims2 partial10 supported

9 copyedit issues flagged: mostly consistency, clarity, grammar.

Reproducibility

6 data/code links checked; 6 live.

datahttps://ngdc.cncb.ac.cn/omix/release/OMIX008341LIVEHTTP 200
Resolves, but the content could not be matched to the paper.
datahttps://clinicaltrials.gov/study/NCT05938920LIVEHTTP 200
Resolves, but the content could not be matched to the paper.
datahttps://clinicaltrials.gov/ct2/show/NCT05154240LIVEHTTP 200
Resolves, but the content could not be matched to the paper.
datahttp://www.chinadrugtrials.org.cn/LIVEHTTP 202
Resolves, but the content could not be matched to the paper.
datahttps://ngdc.cncb.ac.cn/omix/LIVEHTTP 200
Resolves, but the content could not be matched to the paper.
codeGitHubLIVEHTTP 200
https://github.com/HUICUI1992/Code-for-NM
Resolves to GitHub (code repository).

Registered (2 IDs: ClinicalTrials.gov). No reporting guideline cited.

Alpha¹ Rigor Action ItemsFinal pass before journal submission

The published work is largely robust but contains a specific data-integrity concern: the ALT/AE count discrepancy between text and Table 2 is a verifiable internal contradiction that warrants a correction or erratum. An informed reader should weigh this inconsistency, the limited geographical/demographic generalizability, and the absence of a formal power calculation for efficacy. Independent re-analysis would be advisable before relying on the reported effect sizes.

1.
HIGHrigor
Resolve the discrepancy in ALT/AE counts: the text reports '6 in 60 mg QD (33.3%) and 3 in 60 mg QD (16.7%)' but Table 2 shows ALT increase as 0,0,1,0 across groups. Clarify which numbers are correct and publish a correction or erratum if needed. [Results, Primary safety endpoint; Table 2]
Internal contradiction between text and table undermines trust in the reported safety data and must be resolved for readers to interpret the paper.
2.
HIGHreporting
Add the names of the specific institutional review boards or ethics committees that approved the study protocol, and include protocol numbers in the Ethics section of the Methods. [Methods, Study Design]
Current generic statement is inadequate per journal reporting standards; named bodies are required for transparency.
3.
HIGHstatistics
Add version numbers for all software and R packages used in the analysis (e.g., R version, lme4 version, clusterProfiler version) in the Statistical Analysis section. [Methods, Statistical Analysis]
Missing version numbers reduces reproducibility and is a common reviewer request.
4.
HIGHstatistics
Provide exact p-values (not just threshold-adjusted values) for secondary efficacy endpoints where confidence intervals are already given, or state that the study primarily uses CIs and that exact p-values are available in supplementary materials. [Results, Secondary lung function endpoints]
Inconsistent reporting of exact p-values is a gap; CIs alone are acceptable for the primary analysis but exact values for all prespecified tests enhance completeness.
5.
HIGHreporting
Explicitly reference a reporting guideline such as CONSORT in the Methods and indicate that the trial was reported in accordance with it. [Methods, Reporting guideline]
The Nature Portfolio reporting summary is helpful but a direct mention of the specific guideline (e.g., CONSORT) is expected for a clinical trial.
6.
HIGHstatistics
Report explicit checks of statistical test assumptions (normality, homoscedasticity) for the ANCOVA and mixed models, or state that the methods are robust to these assumptions due to trial design. [Methods, Statistical Analysis]
Assumptions verification was flagged by reviewers and is best practice for transparency.
7.
MEDIUMreporting
Include a justification for the sample size based on a clinically meaningful effect size for the primary efficacy endpoint, or clarify why a formal power calculation was not feasible for this phase 2a trial. [Methods, Statistical Analysis]
The current power analysis based on AE detection is appropriate for safety but efficacy-readers will want this explained.
8.
MEDIUMcopyedit
Fix the Abstract and Results text: remove the redundant 'with and' in 'showed improved FVC, with and +98.4 ml' to read 'showed improved FVC, with an increase of +98.4 ml'. [Abstract; Results, Secondary lung function endpoints]
Grammatical error undermines clarity of the main result.
9.
MEDIUMcopyedit
Standardise spacing around 'n=' throughout (e.g., 'n=13/18' vs. 'n = 13/18'); pick one format and apply consistently. [Abstract and throughout]
Inconsistent formatting is a minor readability issue but reflects poor copyediting.
10.
LOWcopyedit
Clarify that the PK evaluable population had smaller N (e.g., n=11) than the total arm size (n=18) due to missing measurable concentrations, and state this explicitly in the text. [Results, Pharmacokinetics]
The text should explain the reduced sample size to avoid confusion.
11.
LOWcopyedit
Add a comma before 'and' in compound sentences where two independent clauses are joined (e.g., in the patients flow paragraph). [Results, Patients, paragraph 1]
Missing comma is a minor punctuation error.

Full text (56,391 chars)Analyzed: Jul 5, 2026

Αlpha¹ assesses methodological rigor, not the importance of the findings. See how we calibrate →