Using Artificial Intelligence Algorithms to Develop Target-specific Ligands

Abstract

Purpose: Pharmaceutical research has recently taken advantage of the rapid advancement in artificial intelligence (AI). The purpose of this study is to use AI to facilitate target-specific ligands development in drug discovery. Here, we used a machine learning algorithm to identify target-specific features of compounds to the metabotropic G-protein coupled receptors 2 and 3 (mGlu2 and mGlu3), which have been targeted for treatment of CNS disorders. Methods: A convolutional neural network (CNN) with 3 hidden layers was made using Tensorflow. We obtained data sets of 315 mGlu2 and 118 mGlu3 compounds and split them into testing and training sets. 75% of each data set was used for training and the remaining was used for testing. We fed the data sets into the CNN and ran the program over 100 iterations with each data set. Results: The neural network was able to differentiate between mGlu2 and mGlu3 compounds in the testing data set with up to a 99.1% accuracy. Visualization of the hidden layers revealed areas in the 2D images that the CNN viewed as important to distinguish the compounds. These identified chemical features can be considered as the target-specific features of the compounds. Conclusions: The neural network is able to differentiate mGlu2 and mGlu3 compounds by 2D representation alone and provide insight to distinguish target-specific features of the compounds through hidden layer visualization. Further testing through cross validation and introduction of a control data set with compounds that bind to neither receptor is needed.