I am currently pursuing my PhD in the Dept. of Pharmaceutical Technology at Jadavpur University, Kolkata, India, with a research focus on computational drug discovery, cheminformatics, quantitative structure-activity relationship (QSAR) modelling, and artificial intelligence (AI)- and machine learning (ML)-based predictive toxicology. My research interests particularly involve the development of reliable, interpretable, and mechanistically informed computational models to predict the biological activities and potential toxicological liabilities of various chemical compounds. The central theme of OpenTox 2026, “Trusted Human-AI Collaboration for Next Generation Risk Assessment and Safe and Sustainable Innovation”, closely aligns with my area of interest. I am particularly enthusiastic about the opportunity to present my research through a poster presentation on an international, open scientific platform, where I can share my findings, receive constructive feedback from experts, and engage with other researchers. Such interactions would significantly contribute to refining my research approaches and strengthening their relevance to real-world toxicological decision-making.
OpenTox Virtual Conference 2026
Machine learning-guided prediction of Hepatobiliary transporters BSEP, MRP2, MRP3 and MRP4 inhibition: Addressing class imbalance and deciphering structural determinants for early assessment of hepatic toxicity
Indrasis Dasgupta1, Shovanlal Gayen1*
1Laboratory of Drug Design and Discovery, Department of Pharmaceutical Technology, Jadavpur University, Kolkata 700032, India
Drug-induced hepatic injury represents a major concern in toxicological safety assessment and pharmaceutical development, frequently contributing to drug development failure, regulatory restrictions, and post-marketing withdrawal. Among the various mechanisms, disruption of hepatobiliary transport processes plays a critical role, particularly in the development of cholestatic liver injury. Hepatobiliary transporters maintain hepatic homeostasis by regulating the uptake, intracellular disposition, and elimination of bile acids, xenobiotics, and their metabolites. The bile salt export pump (ABCB11) mediates canalicular bile acid efflux, whereas multidrug resistance-associated protein 2 (ABCC2) facilitates the biliary excretion of various organic anions and conjugated metabolites. Conversely, MRP3 (ABCC3) and MRP4 (ABCC4), located on the basolateral membrane, provide alternative elimination pathways that may protect hepatocytes during impaired biliary excretion. Inhibition of these transporters can disturb hepatic detoxification mechanisms, promote intracellular accumulation of potentially harmful substances, and increase susceptibility to hepatotoxicity. Therefore, early identification of transporter inhibitors is essential for understanding transporter-mediated toxicity and prioritizing chemicals for further safety evaluation. Consistent with the 3R principles of Replacement, Reduction, and Refinement, computational toxicology offers a promising alternative for supporting hazard identification while minimizing experimental animal use. The present study employs an integrated ML-based computational framework to predict the inhibitory potential of chemical compounds against these four key hepatobiliary transporters. A major methodological challenge of this study was the substantial class imbalance between inhibitors and non inhibitors, which could introduce prediction bias toward the majority class and reduce the identification of minority-class compounds. To address this limitation, different data-balancing approaches, including SMOTE and GHOST, were investigated in conjunction with multiple ML-based classification algorithms. Comparative modelling strategies were employed to examine the influence of class balancing on model discrimination and predictive reliability. The final models were selected by considering both training performance and predictive capability on independent test sets, thereby emphasizing robustness and generalizability rather than apparent model fitting alone. Mechanistic interpretation of the best-performing models was subsequently undertaken to identify molecular characteristics and structural features governing these transporter inhibitions. This approach may support the identification of potential structural alerts, facilitate early-stage hepatotoxicity screening, and guide the design of safer chemicals while contributing to the advancement of animal-alternative toxicological assessment.