PolyCheck: A hybrid model for predicting polypharmacy-induced adverse drug reactions in tuberculosis treatment using heterogenous drug-target-ADR networks

Abstract

Polypharmacy during tuberculosis (TB) treatment, particularly in patients with comorbidities such as diabetes mellitus (DM), significantly increases the risk of adverse drug reactions (ADRs) due to complex drug–drug interactions (DDIs). Existing computational methods primarily focus on pairwise drug interactions, often failing to capture the multifactorial nature of ADRs in polypharmacy contexts. To address this gap, we developed PolyCheck, a hybrid predictive model that integrates network-based and rule-based methods to identify potential ADRs arising from multi-drug regimens. We constructed a heterogeneous Drug–Target–ADR interaction network comprising first-line anti-TB and antidiabetic drugs, their targets, and associated ADRs. The Random Walk with Restart (RWR) algorithm was employed to rank ADR nodes, and a rule-based layer further refined predictions by incorporating the biological relevance of Drug–Target–ADR associations. Evaluation using cross-validation and case-based testing demonstrated strong predictive performance, with accuracy, precision, recall, F1-score, and AUPRC values of 0.70, 0.74, 0.92, 0.81, and 0.74, respectively. PolyCheck offers a scalable and interpretable approach for predicting ADRs in complex treatment regimens and can support safer, individualized TB therapy in patients with comorbid conditions.