LSTM Molecular Descriptor-Free QSAR Application Research Based on Genetic Algorithm Optimization

Traditional machine learning methods, especially random forests and support vector machines, can achieve good prediction accuracy in activity prediction, but the workload of screening molecular descriptors is too large and complicated. Therefore, this paper constructs and uses the QSAR model directly from the SMILES linear text representation of the compound, and uses the genetic algorithm-optimized long-short-term memory neural network model (GA-LSTM) to conduct quantitative structure-activity relationship analysis to predict human protein activity. The experimental results show that the accuracy and efficiency of the QSAR model without molecular descriptors based on GA-LSTM is higher than those of traditional machine learning methods, which provides a new idea and method for the pre-validation of drug development.