Research on GPCR-Ligand Binding Affinity Prediction Models Based on GA-SVR and RF

Abstract

G protein-coupled receptors (GPCRs) are a class of cell surface receptors regulating signal transduction and represent important drug targets. Traditional experimental methods face bottlenecks such as being time-consuming and costly. This study aims to construct high-precision prediction models for GPCR-ligand binding affinity. Utilizing support vector machines optimized by a genetic algorithm and random forest models, based on 933 sets of GPCR-ligand binding data, molecular fingerprints and protein sequence features were extracted. After normalization, the data was partitioned into training, test, and validation sets in an 8:1:1 ratio. Model performance was evaluated using five-fold cross-validation. Experimental results demonstrated that GA-SVR outperformed RF in terms of RMSE and MAE metrics, while RF performed better in R². GA-SVR achieved RMSE, R², MAE values of 0.3388±0.0115, 0.8378±0.0126, and 0.2137±0.0054, respectively. RF achieved RMSE, R2, MAE values of 0.4360±0.0192, 0.9478±0.0042, and 0.3172±0.0110, respectively. GA-SVR showed greate r stability in error control, making it suitable for fine-grained prediction, while RF, due to its powerful nonlinear fitting capability, performed better in capturing overall trends. This study effectively overcomes the efficiency limitations of traditional experimental methods in drug screening, significantly enhances lead compound screening efficacy, and provides innovative solutions for advancing drug discovery and desig.