In silico screening of new inhibitors targeting Aβ42 aggregates using large scale machine-learning and molecular dynamics simulations

Alzheimer's disease (A.D.) is ranked as the third leading cause of death in developed countries and is characterized by the typical presence of senile plaques and neurofibrillary tangles within the brain, usually caused by beta-amyloid like Aβ40 and Aβ42. Aβ42 aggregates more readily, exhibits higher neurotoxicity than Aβ40 and is more significant in the development of A.D. Nowaday, design of new drugs is a costly and time-consuming process, thus it needs in silico approaches to reduce the costs. Here, we used state-of-the-art machine learning (M.L.) methods including random forests (R.F.), support vector machine (SVM), Gaussian Naïve Bayes (GNB), Multilayer perceptron (MLP), logistic regression (L.R.), Gradient boosting (G.B.), extra trees (E.T.) and K-nearest neighbor (KNN) combined with molecular simulations to find novel inhibitors against the Aβ42. The models were trained based on the previously available drugs from BindingDB. A total of 28 out of 1012 phytochemicals from the South African Natural Compounds database (SANCDB) were predicted as active against the amyloid beta 42 by the R.F. model. Virtual screening of the ZINC database revealed that out of the 12000 compounds, a total of 26 compounds were found active by the R.F. model. The 28 hits from the SANCDB and 26 hits from the ZINC database were docked into the model of Aβ42. Additionally, a 200 ns molecular dynamics simulation was performed on the top four docking complexes and the reference complex with WGX-50 bound to assess the binding of the compounds. As compared to the reference compound WGX-50, the four compounds SANC00672, SANC00552, ZINC04234505, and ZINC73633247 showed great binding stability during 200 ns molecular dynamics (M.D.) simulations, which were further validated by determining the best binding free energy (BFE) of the shortlisted hits. To the best of our knowledge, this is the first work to systematically identify novel, potentially effective inhibitors for the Aβ42 using in silico methods, including machine learning-based virtual screening, molecular docking, and molecular dynamics simulations together.