Study of Performance Differences in Small Molecular-Protein Affinity Prediction Based on Multiple Machine Learning Models

The background of this study is that in the field of drug development, there is a high demand for small molecule and protein affinity prediction, and the traditional biochemical assay consumes a large amount of resources and the efficient method of machine learning has gradually become the mainstream method. This study performed three kinds of feature extraction on an identical dataset and used various machine learning models for each of the three feature sets to analyze the properties of protein-small molecule pairs to provide accurate predictions. The project uses linear regression, Multi-layer Perceptron (MLP), Short-Term Memory (LSTM), Recurrent Neural Network (RNN), XGBoost, LightGBM, and Random forest models. The goal is to achieve the most accurate prediction by comparing the performance of different models while reducing resources and computation time. After verification, the LSTM model has the best performance and consumes less resources, which provides a feasible idea for the research in this field.