An innovative machine learning-based QSAR approach for prediction and structural analysis of novel/repurposed acid ceramidase (ASAH1) inhibitors for glioblastoma therapy

Abstract

Acid ceramidase (ASAH1), a lysosomal enzyme that regulates ceramide and sphingosine-1-phosphate balance, has emerged as a promising therapeutic target in Glioblastoma. Inhibiting ASAH1 by carmofur elevates ceramide levels, inducing apoptosis in Glioblastoma cells. However, its clinical application is limited by its instability & toxicological concerns, thereby necessitating the search for more effective inhibitors. We employed an innovative machine learning-quantitative structure–activity relationship (ML-QSAR) approach to investigate & identify related bioactive ASAH1 inhibitors. Herein, we report the results of ML-QSAR modeling utilizing a filtered dataset of 103 inhibitors from ChEMBL & 431 3D descriptors. Multiple algorithmic steps, incorporating top five ML models, were implemented. Among these, our tuned extra trees regressor (ETR) model achieved the highest predictive performance (R2 = 0.867, RMSE = 0.248). Q2(LOO) & Q2(LMO) demonstrated 79.22% & 76.92% (Q2 > 0.5) of inhibitors to be well-predicted, respectively. Descriptor ablation studies identified radial distribution function 20s (RDF20s) and SHAP analysis further demonstrated RDF20s, DPSA-1 & TDB2p as the key structural & pharmacological features. Utilizing this ML-QSAR model, a virtual screening identified 77 promising candidates with N-hexylsalicylamide as the top-most candidate in the ranked list, with superior ADME/T and pharmaco-kinetic characteristics. Notably, Cys143, the key active site residue essential for carmofur interaction, was also observed to be in contact with carbonyl group of N-hexylsalicylamide. MM/PBSA-derived BFE calculations from MD simulations showed that N-hexylsalicylamide had higher negative BFE than carmofur. Following SHAP analyses-based mechanistic interpretations, structural modifications of selected inhibitors led to the design of novel analogs for further testing.