Towards Drug Discovery: A Comparative Study of Machine Learning-enhanced QSAR Prediction

Abstract

In the field of drug discovery, the accurate prediction of bioactive molecules' interactions with biological targets is a significant challenge, limited by the predictive accuracy and handling of complex data in traditional Quantitative Structure-Activity Relationship (QSAR) models. Therefore, our study introduces an innovative approach that integrates advanced machine learning (ML) techniques with QSAR modeling, offering a solution to these limitations. We conducted a comprehensive comparative analysis of various ML algorithms, including decision trees, random forests, support vector machines, deep learning, and ensemble methods, assessing their effectiveness in enhancing QSAR predictions. Our results demonstrate notable improvements in predictive accuracy and efficiency, highlighting the potential of ML-enhanced QSAR models especially with tree-based models in drug discovery. This study contributes significantly to the field by providing a detailed comparison of ML algorithms for QSAR modeling and paving the way for more efficient and accurate drug discovery processes.