Matrix Convolution of Multiple Docking Energy Spectrum for Neural Network Modeling of Multi-Target Pharmacological Activity of Chemical Compounds

An algorithm for matrix convolution of the energy spectrum of multiple docking of ligands into relevant target proteins is presented. Using anxiolytic activity as an example, a matrix convolution of the energy spectrum of multi-target multiple docking into 22 convolutional variables was performed. Using the methods of ANOVA, discriminant analysis and neural network analysis, high statistical significance of using the obtained convolutional variables for assessing the integral multi-target affinity of ligands to a set of relevant biotargets was shown. As a result, a new method for constructing convolutional neural networks for in silico search for pharmacologically active substances based on matrix convolution of the energy spectra of multi-target multiple docking was developed.