Abstract
Recently, machine learning (ML) has gained popularity in the early stages of drug discovery. This trend is unsurprising given the increasing volume of relevant experimental data and the continuous improvement of ML algorithms. However, conventional models, which rely on the principle of molecular similarity, often fail to capture the complexities of chemical interactions, particularly those involving activity cliffs (ACs)─compounds that are structurally similar but exhibit evidently different activity behaviors. In this study, we explore whether transfer learning from AC prediction can enhance prediction of interactions between drug-like compounds and protein targets. We develop a universal model for AC prediction and investigate its impact when transferring learned representations to DTI prediction. Our results suggest that AC-informed transfer learning has the potential to improve the handling of challenging AC-related scenarios, while maintaining overall predictive performance. This study contributes to the ongoing exploration of strategies to enhance ML-based DTI prediction, particularly in cases where conventional approaches face limitations.