Evidential deep learning-based drug-target interaction prediction.

Area of Science:

• Computational chemistry
• Pharmacology
• Artificial intelligence in drug discovery

Background:

• Drug-target interaction (DTI) prediction is vital for drug discovery.
• Current deep learning models for DTI prediction face challenges in uncertainty quantification, robustness, and overconfidence.

Purpose of the Study:

• To introduce EviDTI, a novel approach for uncertainty quantification in DTI prediction using evidential deep learning (EDL).
• To enhance the reliability and robustness of DTI prediction models.
• To accelerate drug discovery by prioritizing high-confidence predictions.

Main Methods:

• EviDTI integrates multiple data dimensions: drug 2D/3D structures and target sequences.
• Utilizes evidential deep learning (EDL) for uncertainty estimation in neural network predictions.
• Evaluated on three benchmark datasets against 11 baseline models.

Main Results:

• EviDTI demonstrates competitive performance compared to existing DTI prediction models.
• The method effectively calibrates prediction errors, providing reliable uncertainty estimates.
• Uncertainty-guided predictions successfully identified novel potential modulators for tyrosine kinases FAK and FLT3 in a case study.

Conclusions:

• Evidential deep learning (EDL) offers a robust framework for uncertainty quantification in DTI prediction.
• EviDTI enhances the efficiency and reliability of the drug discovery pipeline.
• This approach has significant implications for accelerating the identification of novel drug candidates.