pDILI_v1: A Web-Based Machine Learning Tool for Predicting Drug-Induced Liver Injury (DILI) Integrating Chemical Space Analysis and Molecular Fingerprints

Abstract

Drug-induced liver injury (DILI) represents a critical safety concern for drug development, regulatory oversight, and clinical practice, with substantial economic and public health implications. While predicting DILI risk in humans has garnered significant attention, the associated chemical space has remained insufficiently explored. This study addresses this gap through a comprehensive computational approach, leveraging machine learning (ML) to investigate structural determinants of DILI risk systematically. The study focuses on three key objectives: (i) exploring the chemical space and scaffold diversity associated with DILI; (ii) employing fragment-based approaches to identify structural alerts (SAs) that influence DILI risk; and (iii) developing supervised ML models to not only predict DILI risk but also elucidate the structural significance of molecular fingerprints. To broaden accessibility, we introduce pDILI_v1, a Python-based web application available at https://pdiliv1web.streamlit.app/. This user-friendly platform facilitates the prediction and visualization of DILI risk, enabling both experts and nonexperts to screen compounds effectively. Additional formats, including a Google Colab notebook and a graphical user interface (GUI) for Windows, ensure flexibility for diverse user needs. The proposed models demonstrate the potential for early identification of hepatotoxic risks in drug candidates, providing critical insights into drug discovery and development. By integrating ML-driven predictions with chemical space analysis, this research advances the field of drug safety evaluation, contributing to the development of safer pharmaceuticals and mitigating the risks of DILI.

Figure 1

Workflow of the current study involves different approaches such as (i) chemical space analysis, (ii) the fragment-based approach, and (iii) ML-based QSAR.

Figure 2

Bin plots of each feature (Class, LogP, MW, nAR, HBA, HBD, nRings, nRB, and TPSA), colored by DILI Toxic (1) and DILI Nontoxic (0).

Figure 3

Representative drug compounds with DILI toxic (T) and DILI nontoxic (N) fingerprints/substructural features. DILI toxic fragments promote DILI risk, while DILI nontoxic fragments hinder DILI risk. These substructural features were produced by the ECFP-6 fingerprint descriptor.

Figure 4

68 descriptors with their importance score.

Figure 5

Heatmap of the correlation matrix of the selected descriptors.

Figure 6

(a) Optimization history plot of the hyperparameter optimization process, (b) Slice plot of specific hyperparameters (max_depth, min_samples_leaf, min_samples_split, and n_estimators) with respect to the objective value.

Figure 7

Partial dependence plot (PDP) of descriptors (a) SLogP, (b) GATS 2d, and (c) AATS0v. Two-variable PDP of chemical descriptors (d) SLogP vs GATS 2d and (e) SLogP vs AATS0v. The contour plot uses color gradients to represent regions corresponding to specific numerical values, as indicated by the contour levels. These values likely reflect a performance metric with the highest values appearing in the yellow-green regions and lower values in the darker blue and purple areas. The gradient transitions from darker to lighter shades, where lighter regions correspond to higher values.

Figure 8

Mechanistic interpretations of the descriptors (SLogP and GATS 2d) and mathematical contributions to the ML model.

Figure 9

Mechanistic interpretations of the descriptors (SLogP and AATS0v) and mathematical contributions to the ML model.

Figure 10

Applicability domain (AD) was based on the leverage approach. The outliers (those outside AD) identified by leverage are highlighted in blue circles.