AI/ML-Driven DPP-4 Inhibitor Predictor (d4p_v1) for Enhanced Type 2 Diabetes Mellitus Management: Insights Into Chemical Space, Fingerprints, and Electrostatic Potential Maps

Abstract

Dipeptidyl peptidase-4 inhibitors (DPP-4i) represent a relatively new class of oral antidiabetic drugs. This study focuses on: (a) identifying favourable and unfavourable fingerprints governing DPP-4 inhibition using fragment-based analysis, (b) validating key fingerprints through HOMO–LUMO gap analysis and electrostatic potential (ESP) maps, and (c) developing AI/ML-driven DPP-4 predictor, an online cheminformatics tool for efficient DPP-4i screening using a trained, validated AI/ML model. The fragment-based QSAR model finds key substructures linked to potent DPP-4 inhibition, including 2-cyanopyrrolidine, 3-amino tetrahydropyran, and difluoro phenyl groups. D0010 (3-aminotetrahydropyran fingerprint G10) is the most reactive, while D0094 (difluorophenyl fingerprint G14) is the most stable, with D0012 and D0013 (2-cyanopyrrolidine fingerprints G1, G5) offering a balance between stability and reactivity. In addition, the d4p_v1 tool (https://github.com/Amincheminfom/d4p_v1) reliably distinguishes active and inactive DPP-4i using molecular descriptors derived from input SMILES strings. Therefore, this study not only revealed the chemical space of DPP-4i but also opened up a horizon in developing novel potent DPP-4i for the management of type 2 diabetes mellitus (T2DM) in the future.