pDILI_v1: An Open-Source Web-Based Tool for Predicting Drug-Induced Liver Injury (DILI)

Supratik Kar

 Co-Presenters: Individual Presentation
College: Dorothy and George Hennings College of Science, Mathematics and Technology
Department: Chemistry & Physics
 

Abstract:

Drug-induced liver injury (DILI) presents a major challenge in drug development and regulatory assessment. While machine learning (ML)-based approaches for DILI risk prediction are gaining traction, the underlying chemical space remains underexplored. To address this, we introduce pDILI_v1, a Python-based platform designed for DILI risk assessment using molecular fingerprints and ML models. The tool integrates (i) chemical space exploration, (ii) scaffold and fragment-based structural alerts, and (iii) supervised ML models to provide a comprehensive risk evaluation. pDILI_v1 is available in multiple formats to ensure accessibility across diverse user environments. The primary interface is a Streamlit-based web application (https://pdiliv1web.streamlit.app), where users can input a SMILES string to predict DILI risk and visualize the query compound’s position within the applicability domain. A Google Colab notebook is also provided, requiring dataset uploads from the associated GitHub repository (https://github.com/Amincheminfom/pDILI_v1). Additionally, a standalone GUI version is available for Windows via Anaconda, enabling offline use. By integrating ML-driven predictions with chemical space analysis, pDILI_v1 facilitates early hepatotoxicity screening, enhancing drug safety assessment and accelerating pharmaceutical development.