A computational and in silico study of machine learning-assisted design of boron and selenium compounds against Alzheimer's disease targets

Alzheimer's disease (AD) is the most common neurodegenerative disease worldwide with over 55 million people affected and projected to reach more than 130 million by 2050. Although substantial research is underway, FDA approved therapies are palliative only. In this work, we employed machine learning (ML) algorithms, including Random Forest, Support Vector Machine, Gradient Boosting (XGBoost), Deep Neural Networks, and QSAR multiple linear regression to screen and design new boron-selenium (BSe) hybrid compounds as potential inhibitors for acetylcholinesterase (AChE), butyrylcholinesterase (BuChE), betasecretase 1 (BACE1) and tau protein aggregation. A library of 3562 heteroatom scaffolds including boron pinacolates, organoboron-selenides and ebselen-boronate hybrids was screened. The ML models predicted targets with R^2 of 0.856–0.921. The top-ranked compounds (BSe-01 to BSe-05) exhibited docking energies ranging from −9.78 to −12.15 kcal/mol at AChE, which were higher than that of donepezil (−10.80 kcal/mol). Full ADMET profiling confirmed blood brain barrier permeability, high gastrointestinal absorption and negligible toxicity. The predicted pIC_50 values of 6.98–7.63 suggest sub-micromolar potencies. Molecular dynamics simulations over 100 ns showed stable protein–ligand complexes. These results suggest that boron-selenium hybrids are a promising next generation scaffold for the development of multi-target anti-Alzheimer drugs.