Towards a more general drug target interaction prediction model using transfer learning.

Area of Science:

• Computational chemistry
• Bioinformatics
• Artificial intelligence

Background:

• Drug-target interaction (DTI) is crucial for discovering new disease treatments.
• Artificial intelligence (AI) is increasingly applied to DTI prediction.
• Transfer learning's potential in DTI prediction remains underexplored.

Purpose of the Study:

• To develop a more generalizable DTI prediction model using transfer learning.
• To investigate the effectiveness of CNN, RNN, and Transformer models in DTI prediction via transfer learning.

Main Methods:

• Tested CNN, RNN, and Transformer models for DTI prediction.
• Utilized two public DTI datasets: BindingDB and DAVIS.
• Implemented transfer learning by pre-training models on one dataset and fine-tuning on another.

Main Results:

• The CNN model pre-trained on BindingDB demonstrated superior performance.
• Fine-tuning the BindingDB pre-trained model on the DAVIS dataset resulted in a 6% increase in AUPRC compared to models without pre-training.

Conclusions:

• A CNN model pre-trained on BindingDB is recommended for general DTI prediction tasks.
• Transfer learning significantly boosts DTI prediction performance, especially when adapting models to new biological contexts.