MoDTI: Modular Framework For Evaluating Inductive Biases in DTI Modeling

Roy Henha Eyono, Prudencio Tossou, Cas Wognum, Emmanuel Noutahi

Keywords: Modularity, Multi-View, DTI, Neural Architecture, Inductive Biases

TL;DR: We propose a modular framework called MoDTI that allows us to jointly investigate three (3) inductive biases: protein representation, multi-view learning, modularity.

Abstract: Drug-Target Interaction (DTI) prediction is a critical problem in drug discovery, and machine learning (ML) has shown great potential in feature-based DTI prediction. However, selecting an appropriate ML architecture from the vast number of available biomolecular representations is challenging. To address this issue, we propose MoDTI, a modular framework that enables the exploration of three key inductive biases in DTI prediction: protein representation, multi-view learning, and modularity. We evaluate the impact of each inductive bias on DTI prediction performance and compare the performance of MoDTI against existing state-of-the-art models on multiple benchmarks. Our experiments with MoDTI provide valuable insights into the role of modularity, capacity, representation redundancy, and orthogonality in terms of generalization and interpretability. They also enable the provision of general guidelines for the rapid development of more accurate DTI models.