ABSTRACT Anti-TNFα therapies are widely used to treat autoimmune diseases, including inflammatory bowel disease (IBD), a chronic inflammation of the digestive system which presents both an economic burden as well as prominent disability and morbidity. These therapies, however, are largely ineffective in IBD patients, with primary non- response rates varying from 10-30%, and with 23-46% of patients losing response over time. They also lead to adverse health outcomes including increased risk of infections (most notably reactivation of tuberculosis), liver problems, arthritis, and lymphoma. Due to biologic switching where multiple drugs are tried in sequence until an effect is observed, they often lead to overall increased healthcare costs. There is no clear predictive factor of response or loss of response to anti-TNFα therapies, with current research focusing on efficacious dosage and inhibitor selection. While numerous predictors of response have been identified in studies, none are robust enough to impact clinical practice. Overcoming this gap in the administration of anti-TNFα therapies will be an important next step in improving their utility and reducing overall healthcare costs, morbidity, and mortality. This project proposes a multicohort meta-analysis and machine learning approach to discover and validate a prognostic gene signature that can differentiate anti-TNFα responders from non-responders in IBD patients, allowing for improved decision-making and positive health outcomes. Despite the broad biological, clinical, and technical heterogeneity inherent in such a task, the Inflammatix analytical framework was able to identify significant differentially expressed genes that discriminate anti-TNFα responders from non-responders (mean discovery AUC of ~0.83 and mean validation AUC of ~0.81). Supported by these preliminary results, this project will (1) discover a robust, clinically translatable multi-gene prognostic signature using publicly available datasets, (2) generate independent gene expression data from retrospective cohorts of mucosal biopsy samples (n=150) to validate the signature (target AUC > ~0.8), and (3) use our proprietary Inflammatix machine learning (IML) platform to train and validate a plethora of machine learning algorithms to develop a robust, generalizable classifier (target AUC ~ 0.8 - 0.9) that is ready for clinical validation via prospective studies. Our novel prognostic signature will transform the clinical paradigm in the use and administration of anti-TNFα therapies, maximizing treatment benefit while minimizing patient exposure to potentially harmful side effects and adverse health outcomes, consequently reducing financial burden for the healthcare system and patient.