Ensemble Learning, Deep Learning-Based and Molecular Descriptor-Based Quantitative Structure–Activity Relationships

Abstract

A deep learning-based quantitative structure–activity relationship analysis, namely the molecular image-based DeepSNAP–deep learning method, can successfully and automatically capture the spatial and temporal features in an image generated from a three-dimensional (3D) structure of a chemical compound. It allows building high-performance prediction models without extracting and selecting features because of its powerful feature discrimination capability. Deep learning (DL) is based on a neural network with multiple intermediate layers that makes it possible to solve highly complex problems and improve the prediction accuracy by increasing the number of hidden layers. However, DL models are too complex when it comes to understanding the derivation of predictions. Instead, molecular descriptor-based machine learning has clear features owing to the selection and analysis of features. However, molecular descriptor-based machine learning has some limitations in terms of prediction performance, calculation cost, feature selection, etc., while the DeepSNAP–deep learning method outperforms molecular descriptor-based machine learning due to the utilization of 3D structure information and the advanced computer processing power of DL.

Active and machine learning-enhanced discovery of new FGFR3 inhibitor, Rhapontin, through virtual screening of receptor structures and anti-cancer activity assessment

“…The present study presents a systematic exploration aimed at augmenting the efficacy of Ceritinib, a prominent FGFR3 inhibitor, via the integration of natural compound-derived alternatives. Our investigation embraces a multidimensional approach, employing active learning derived virtual screen ( Ma et al, 2009 ), deep learning derived QSAR modeling ( Matsuzaka and Uesawa, 2023 ), molecular dynamics simulations ( Duay et al, 2023 ), and biological assays to dissect the mechanisms underlying the potential synergy between Ceritinib and the identified natural compounds.…”

Excuse me, there is a mutant in my bioactivity soup! A comprehensive analysis of the genetic variability landscape of bioactivity databases and its effect on activity modelling

“…Quantitative structure-activity relationship (QSAR) modelling has been around for a long time and can be used to predict ligand bioactivity for a target of interest based on the compound's chemical structural characteristics 3 . Over time other bioactivity prediction strategies have emerged that include information other than chemistry-derived features [4][5][6][7][8] . An example is proteochemometric (PCM) modelling, where the protein characteristics are considered in addition to ligand molecular structure, allowing for bioactivity predictions on several targets simultaneously [8][9][10] .…”

MATH: A Deep Learning Approach in QSAR for Estrogen Receptor Alpha Inhibitors

“…Nowadays, many QSAR developments apply a multi-objective QSAR approach to drug discovery [11]. Traditional QSAR methods have transitioned towards machine learning (ML) models, including deep learning (DL) models, to achieve more diverse variations in the resulting predictors.…”