Transfer learning for drug-target interaction prediction.

Area of Science:

• Computational biology
• Artificial intelligence in drug discovery
• Bioinformatics

Background:

• Drug-target interaction (DTI) prediction is crucial for drug discovery.
• AI-driven DTI prediction methods require substantial training data, which is often unavailable for understudied proteins.
• Deep transfer learning offers a potential solution for DTI prediction with limited data.

Purpose of the Study:

• To investigate the efficacy of deep transfer learning for predicting drug-target interactions (DTI) involving understudied proteins with scarce training data.
• To evaluate the performance of transfer learning compared to training deep neural networks from scratch for DTI prediction.

Main Methods:

• A deep neural network classifier was pre-trained on a large, generalized source dataset.
• The pre-trained network was fine-tuned using smaller, specialized target datasets for understudied protein families (e.g., transporters, nuclear receptors).
• Performance was systematically evaluated across different transfer learning strategies and compared to conventional training from scratch.

Main Results:

• Deep transfer learning significantly outperformed training from scratch when the target training dataset contained fewer than 100 compounds.
• The study demonstrates the advantage of transfer learning for predicting drug binders to understudied protein targets.
• The approach was validated using critical protein families in biomedicine, including kinases, GPCRs, ion channels, nuclear receptors, proteases, and transporters.

Conclusions:

• Deep transfer learning is a powerful and advantageous approach for drug-target interaction prediction, especially for targets with limited available training data.
• This method can accelerate the identification of drug candidates for understudied proteins, addressing a key bottleneck in drug discovery.
• The developed models and code are publicly available, facilitating further research and application.