Bridging antiviral drug discovery with a large language model-powered framework.

Area of Science:

• Computational chemistry and pharmacology
• Drug discovery and development
• Virology and infectious diseases

Background:

• Viral infections necessitate continuous development of effective antiviral therapies.
• Current computational antiviral drug discovery often overlooks biological phenotypes, focusing narrowly on target binding.
• Existing methods lack comprehensive integration of phenotype-based drug discovery (PBDD) and target-based drug discovery (TBDD).

Purpose of the Study:

• To introduce DeepAVC, a large language model-powered framework for advanced antiviral compound prediction.
• To integrate PBDD (DeepPAVC) and TBDD (DeepTAVC) within a synergistic computational approach.
• To enhance the interpretability and predictive power of computational antiviral drug discovery.

Main Methods:

• Development of DeepAVC, a framework combining DeepPAVC for PBDD and DeepTAVC for TBDD.
• Utilizing large language models to integrate diverse drug discovery strategies.
• Validation through in vitro and in vivo experimental assays.
• Analysis of compound-protein interactions for interpretability.

Main Results:

• DeepAVC significantly outperforms existing computational baselines in predicting antiviral compounds.
• The framework provides high interpretability by identifying critical atoms and residues in drug-target interactions.
• DeepPAVC and DeepTAVC demonstrate complementary and synergistic effects within the DeepAVC framework.
• Identification of MNS as a novel broad-spectrum antiviral compound with superior efficacy compared to Sisunatovir.

Conclusions:

• DeepAVC represents a significant advancement in computational antiviral drug discovery.
• The synergistic integration of PBDD and TBDD enhances prediction accuracy and biological relevance.
• DeepAVC is a valuable and interpretable tool for identifying novel antiviral agents and accelerating drug development.