Cheminformatics and Machine Learning-Driven QSAR Analysis of SPHK2 Inhibitors for Anticancer Drug Design

Sphingosine kinase 2 (SPHK2) plays a pivotal role in sphingolipid metabolism and has emerged as a therapeutic target in cancer due to its involvement in tumor proliferation and resistance mechanisms.

A dataset of 269 SPHK2-targeting compounds from ChEMBL was analyzed using five molecular descriptor sets: PubChem, MACCS, CDK, Substructure, and Klekota- Roth. Six machine learning algorithms were applied to develop QSAR models, and they were validated using ROC-AUC, PCA, R2, and Q2 metrics. Top-ranked compounds were subsequently evaluated using Glide docking and MMGBSA energy calculations.

Random Forest models demonstrated the best predictive performance, especially with Klekota-Roth and PubChem descriptors (R2 = 0.78; Q2 = 0.72). Lead compounds, namely CHEMBL2409888 and CHEMBL2409889, exhibited strong binding affinities (ΔG = -58.02 and -58.13 kcal/mol), interacting with key residues GLU293, ARG652, and HIS289. Inactive compounds showed reduced binding strength and limited residue interactions.

This integrative computational pipeline successfully identified structurally significant SPHK2 inhibitors and highlighted molecular features contributing to activity. While the results offer mechanistic insights and a rational framework for further optimization, findings are based solely on in silico predictions.

This study presents a predictive framework combining machine learning and molecular modeling to identify selective SPHK2 inhibitors, offering valuable candidates for further synthesis and biological validation.