DTI-LM: language model powered drug-target interaction prediction.

Area of Science:

• Computational biology
• Drug discovery and development
• Bioinformatics

Background:

• Drug-target interactions (DTIs) are crucial for drug discovery.
• Sequence-based computational models offer efficient DTI prediction.
• Cold start DTI prediction for unknown drugs or proteins remains a challenge.

Purpose of the Study:

• To introduce DTI-LM, a novel framework for DTI prediction.
• To leverage pretrained language models and neighborhood information for enhanced DTI prediction.
• To bridge the gap between warm start and cold start DTI predictions using sequence representations.

Main Methods:

• Utilized advanced pretrained language models for sequence representation of drugs and proteins.
• Incorporated neighborhood information via a graph attention network.
• Developed DTI-LM framework relying solely on sequence data.

Main Results:

• DTI-LM achieved state-of-the-art performance on DTI prediction tasks across four datasets.
• Demonstrated significant improvements in cold start predictions for proteins.
• Observed a persistent disparity in cold start predictions between proteins and drugs.

Conclusions:

• DTI-LM effectively predicts drug-target interactions using sequence-based language models.
• The framework shows promise in addressing cold start challenges in DTI prediction.
• Further research is needed to address the observed disparities in drug and protein cold start predictions.