What question did this study set out to answer?

The research aims to develop a machine learning model to predict drug efficacy in inflammatory bowel disease using preclinical data and patient transcriptomic data.

January 23, 2026

P0318Facing the valley of death in Inflammatory Bowel Disease: a machine learning approach to predict the clinical efficacy of drug candidates from preclinical data

Key Points

The research aims to develop a machine learning model to predict drug efficacy in inflammatory bowel disease using preclinical data and patient transcriptomic data.
Developed a machine learning model with two modes: Target Mode and Preclinical-reinforced Mode
Utilized transcriptomic data from IBD patients and a murine model
Compared predicted drug responses with observed clinical outcomes in TNFi-treated patients
Model predictions matched observed remission rates closely for UC and CD patients
In Target Mode, the model achieved an AUROC of 0.63 for both UC and CD
In Preclinical-reinforced Mode, AUROC improved to 0.65 for UC and 0.72 for CD
High specificity (82%) and NPV (78%) were obtained, indicating effective identification of non-responders

Abstract

Abstract Background Translation from preclinical to early-phase clinical research is a critical step of drug development, with up to 70% of drug candidates failing to show clinical efficacy despite promising preclinical results 1,2. Indeed, in IBD, current preclinical models often fail to reproduce the chronic, heterogeneous, and systemic nature of disease. To address this gap, we developed a machine learning model that operates in 2 modes. In Target Mode, the model is provided with a therapeutic target and patient transcriptomic data, and it predicts patient-level drug response. In Preclinical-reinforced Mode, predictive performances can be enhanced by incorporating omics data from preclinical animal models. In this study, we retrospectively validated the model by comparing predicted effects of TNFi drugs with observed clinical responses of TNFi-treated IBD patients. Methods Two scenarios were studied. In Scenario A (Target Mode), the model used TNFSF2 as a drug target to predict patient-level endoscopic remission. In Scenario B (Preclinical-reinforced Mode), additional preclinical data, including transcriptomic data from a murine IBD model, were incorporated. The data flow is presented in Fig 1. In both scenarios, the model used patient transcriptomic data (colon/ileum RNAseq) from IBD patients prior to anti-TNF therapy, and predictions were compared with observed endoscopic outcomes. Performance was assessed using AUROC, sensitivity, specificity, PPV, and NPV. Ultimately, patient-level predictions were aggregated to compare with cohort-level remission rates. Results The cohort included 26 UC (9 responders, 35%) and 17 CD patients (9 responders, 53%) 3. Aggregated predictions closely matched observed remission rates (predicted vs. observed: UC 31% vs. 35%; CD 59% vs. 53%), suggesting the model can predict remission rates prior to phase 2 trials. We then evaluated the model in two scenarios. In scenario A, provided only with the drug target TNFSF2, the model used baseline transcriptomic data to predict endoscopic remission with an AUROC of 0.63 (95% CI, 0.41–0.84) in UC and 0.63 (95% CI, 0.32–0.90) in CD. In scenario B, when provided with additional preclinical data from a TNFR2 k/o mouse study, AUROC improved to 0.65 (95% CI, 0.38–0.88) in UC and 0.72 (95% CI 0.45–0.95) in CD. In Scenario B, the model achieved high specificity (82%) and NPV (78%) in UC, indicating reliable identification of TNFi non-responders and potential for more efficient phase 2 enrolment (see Tab 1). Conclusion Our model can estimate remission rates before phase 2, enabling early indication prioritization. Also, accurate identification of non-responders would improve patient selection and trial probability of success. Further validation across additional targets is ongoing. Reference: 1. Seyhan, A.A. Lost in translation: the valley of death across preclinical and clinical divide – identification of problems and overcoming obstacles. transl med commun 4, 18 (2019).https://doi.org/10.1186/s41231-019-0050-72. Patel DD et al. Phase 2 to phase 3 clinical trial transitions: Reasons for success and failure in immunologic diseases. J Allergy Clin Immunol. 2017 Sep;140(3):685-687. doi: 10.1016/j.jaci.2017.04.029. Epub 2017 May 12. PMID: 28506849.3. Verstockt B et al. Low TREM1 expression in whole blood predicts anti-TNF response in inflammatory bowel disease. EBioMedicine. 2019 Feb;40:733-742. doi: 10.1016/j.ebiom.2019.01.027.Conflict of interest: Fouché, Aziz: Employee of Scienta Lab Corney, Matthew: Employee of Scienta Lab Marschall, Pierre: Employee of Scienta Lab Strozzi, Francesco: Employee of Scienta Lab Verstockt, Bram: Research support from AbbVie, Biora Therapeutics, Celltrion, Landos, Pfizer, Sanofi, Sossei Heptares/Nxera and Takeda. Speaker’s fees from Abbvie, Agomab, Alfasigma, Biogen, Bristol Myers Squibb, Celltrion, Eli Lily, Falk, Ferring, Galapagos, Materia Prima, Johnson and Johnson, Pfizer, Sandoz, Takeda, Tillots Pharma, Truvion and Viatris. Consultancy fees from Abbvie, Alfasigma, Alimentiv, Anaptys Bio, Applied Strategic, Astrazeneca, Atheneum, BenevolentAI, Biora Therapeutics, Boxer Capital, Bristol Myers Squibb, Domain Therapeutics, Eli Lily, Galapagos, Guidepont, Landos, Merck, Mirador Therapeutics, Mylan, Nxera, Inotrem, Ipsos, Johnson and Johnson, Pfizer, Sandoz, Sanofi, Santa Ana Bio, Sapphire Therapeutics, Sosei Heptares, Takeda, Tillots Pharma and Viatris. Stock options Vagustim and Thethis Pharma. Raine, Timothy: Grant: Abbvie, Takeda Personal Fees: TR has received research/educational grants and/or speaker/consultation fees from Abbvie, Alfasigma, Arena, Aslan, AstraZeneca, Boehringer-Ingelheim, BMS, Celgene, Domain Therapeutics, Eli Lilly, Ferring, Galapagos, Gilead, GSK, Heptares, LabGenius, Janssen, MonteRosa, Mylan, MSD, Novartis, Numab, Pfizer, Roche, Sandoz, Scientia, Takeda, UCB and XAP therapeutics Duquesne, Julien: Employee and shareholder of Scienta Lab Mr. Bouget, Vincent: Employee and shareholder of Scienta Lab.

Mark Helpful

Bookmark

Relay