Enhancing Drug-Target Interaction Prediction through Transfer Learning from Activity Cliff Prediction Tasks.

Area of Science:

• Computational chemistry
• Machine learning in drug discovery
• Bioinformatics

Background:

• Machine learning (ML) is increasingly used in early drug discovery due to data growth and algorithm improvements.
• Conventional ML models often struggle with activity cliffs (ACs)—structurally similar compounds with disparate activities—limiting drug-target interaction (DTI) prediction accuracy.
• Molecular similarity alone is insufficient for capturing complex chemical interaction nuances.

Purpose of the Study:

• To investigate if transfer learning from activity cliff (AC) prediction can enhance drug-target interaction (DTI) prediction.
• To develop a universal model for AC prediction and assess its utility in transfer learning for DTI tasks.
• To address limitations of conventional ML models in handling AC-related scenarios.

Main Methods:

• Development of a universal model for predicting activity cliffs (ACs).
• Application of transfer learning, using representations learned from AC prediction, to drug-target interaction (DTI) prediction tasks.
• Evaluation of the impact of AC-informed transfer learning on DTI predictive performance, particularly for challenging AC cases.

Main Results:

• AC-informed transfer learning shows potential to improve the prediction of drug-target interactions (DTIs).
• The approach demonstrates enhanced capability in managing difficult scenarios involving activity cliffs (ACs).
• Overall predictive performance for DTI was maintained while improving the handling of AC-related complexities.

Conclusions:

• Transfer learning from AC prediction offers a promising strategy to enhance ML-based DTI prediction.
• This method can improve the accuracy and robustness of models dealing with structurally similar compounds exhibiting varied activities.
• The study contributes valuable insights into advanced ML techniques for overcoming limitations in drug discovery prediction.