Unveiling Protein-Ligand Interactions: Regression Methods for Binding Affinity Prediction

Protein-ligand binding affinity is a cornerstone of structure-based drug design. The ability to computationally predict ligand-protein binding affinities represents a significant challenge in computational medicinal chemistry. This capability holds immense value in the early stages of drug discovery, enabling researchers to leverage mathematical models to evaluate protein-ligand interactions. Moreover, advancements in computational methods, particularly machine learning techniques, have revolutionized the prediction of binding affinity, offering a cost-effective and efficient alternative to traditional experimental approaches. This interdisciplinary convergence of bioinformatics and machine learning not only enhances our understanding of molecular interactions but also accelerates the drug discovery process by facilitating the selection and optimization of promising drug candidates. This study explored the use of various regression methods, including linear regression, k-NN, gradient boosting, support vector machines, Random Forest, and multi-layer perceptron algorithms, on existing protein-ligand binding affinity data to predict the binding affinity of unknown ligands. The models were built using the latest dataset, Plas-20k, which covers protein-ligand binding affinities. To evaluate our models, the coefficient of determination (R2), Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) metrics were used. According to experimental results, the Random Forest model performed best, achieving an R2 of 0.9059, an RMSE of 0.8401, and an MAE of 0.6528.