Machine learning–guided optimization of MALT1 inhibitors for diffuse large B-cell lymphoma via QSAR modelling, molecular docking, ADMET profiling, and molecular dynamics simulations
Josiah Joseph Isah, Adamu Uzairu, Sani Uba, Muhammad Tukur Ibrahim
Abstract
Background The paracaspase MALT1(mucosa-associated lymphoid tissue lymphoma translocation protein 1) plays a pivotal oncogenic role in diffuse large B-cell lymphoma (DLBCL). However, many reported inhibitors exhibit suboptimal stability, drug-likeness, and binding persistence. Objective To develop an integrated machine learning and structure-based pipeline to design and optimize potent, synthetically accessible MALT1 inhibitors. Methods Experimentally validated MALT1 inhibitors from ChEMBL (n = 319) were used to build a Random Forest quantitative structure–activity relationship (QSAR) model using RDKit molecular descriptors. Top-ranked candidates were subjected to molecular docking (PDB ID: 6YN8), absorption–distribution–metabolism–excretion–toxicity (ADMET) evaluation, and structure-guided analogue design. Molecular dynamics simulations and Molecular mechanics–generalised Born surface area (MM-GBSA) binding energy analyses were performed to assess conformational stability and protein–ligand interactions. Results The optimized QSAR model achieved strong predictive performance (external R 2 = 0.65). Docking identified compounds 5 and 6 (−9.0 kcal/mol) with hydrogen bonding around LYS360/ALA361 and hydrophobic groove residues. Four generated analogues (5a–5d) showed improved binding, with compound 5a demonstrating the highest affinity (−11.1 kcal/mol), near-complete predicted absorption (∼99 %), synthetic tractability, and improved clearance. MD simulations revealed that 5a maintained lower root-mean-square deviation (RMSD), reduced solvent-accessible surface area (SASA), and stronger hydrogen-bond occupancy compared to parent compound 5. MM-GBSA analysis showed stronger binding free energy for 5a (−86.23 kcal/mol) versus 5 (−77.75 kcal/mol). Conclusion The machine learning–guided integrated framework successfully optimized MALT1 inhibitors, identifying compound 5a as a promising lead scaffold with therapeutic potential to disrupt oncogenic MALT1 signalling in DLBCL, warranting synthesis and experimental evaluation.