TAPB: an interventional debiasing framework for alleviating target prior bias in drug-target interaction prediction

Area of Science:

• Computational Biology
• Pharmacology
• Machine Learning

Background:

• Drug-Target Interaction (DTI) prediction is crucial for drug repurposing.
• Existing DTI models suffer from
• target prior bias,
• stemming from imbalanced target label distributions in training data.

Purpose of the Study:

• To introduce a novel debiasing framework, TAPB, to address target prior bias in DTI prediction.
• To leverage causal inference to compute interventional probabilities P(Y∣D, do(T)) for accurate DTI modeling.

Main Methods:

• Amino acid randomization for feature representation.
• Confounder Alignment Module (CAM) for bias mitigation.
• Interventional training with backdoor adjustment for causal effect estimation.

Main Results:

• TAPB framework effectively alleviates target prior bias.
• Achieved competitive performance compared to existing DTI prediction methods.
• Demonstrated enhanced model generalization and provided interpretable DTI insights.

Conclusions:

• Target prior bias is a significant confounder in DTI prediction.
• TAPB offers a robust solution for accurate and generalizable DTI prediction.
• The framework enhances understanding of drug-target interaction mechanisms.