Large Language Models Enable Semantic Alignment for Cold-Start Compound-Protein Interaction Prediction.

Area of Science:

• Drug Discovery
• Computational Biology
• Bioinformatics

Background:

• Cold-start compound-protein interaction (CPI) prediction is crucial for drug discovery but challenging due to the need to generalize beyond training data.
• Existing methods often fail to capture semantic relationships between compounds and proteins, leading to poor performance in cold-start scenarios.

Purpose of the Study:

• To develop a novel framework, SACS-CPI, for robust cold-start CPI prediction.
• To leverage large language model (LLM)-driven semantic alignment to enhance generalization capabilities.

Main Methods:

• SACS-CPI maps compounds and proteins into a shared semantic space using LLM-driven alignment.
• The framework integrates general semantic knowledge with task-specific features via dynamic gating.
• Bidirectional cross-attention and bilinear pooling capture high-order interaction patterns.

Main Results:

• SACS-CPI significantly outperforms existing state-of-the-art methods on multiple benchmark datasets.
• The model demonstrates robust performance gains in compound cold-start, protein cold-start, and blind-start settings.
• Semantic alignment is shown to be an effective inductive bias for cold-start CPI prediction.

Conclusions:

• LLM-driven semantic alignment offers a powerful approach to address the cold-start problem in CPI prediction.
• SACS-CPI provides a more generalizable and accurate method for identifying potential drug candidates.
• The proposed framework advances the field of computational drug discovery and personalized medicine.