ColdstartCPI: Induced-fit theory-guided DTI predictive model with improved generalization performance.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Predicting compound-protein interactions (CPIs) is vital for drug discovery.
• Traditional rigid docking methods struggle with molecular flexibility and sparse data.

Purpose of the Study:

• Introduce ColdstartCPI, a novel framework for predicting CPIs.
• Address limitations of existing methods, particularly for novel compounds and proteins.

Main Methods:

• Leverage unsupervised pre-training features and a Transformer module.
• Incorporate induced-fit theory to model molecular flexibility.
• Utilize sequence-based learning for compounds and proteins.

Main Results:

• ColdstartCPI outperforms state-of-the-art sequence-based models, especially for unseen targets.
• Demonstrates strong generalization in virtual screening compared to structure-based methods.
• Excels in sparse and low-similarity data conditions.

Conclusions:

• ColdstartCPI offers a flexible, sequence-based approach to drug design.
• Presents a promising tool for drug discovery, particularly in data-limited scenarios.
• Validated through molecular docking and binding free energy calculations.