Predicting and generating antibiotics against future pathogens with ApexOracle.

Area of Science:

• Computational biology
• Drug discovery
• Artificial intelligence in medicine

Background:

• Antimicrobial resistance (AMR) is a growing global health threat, necessitating novel antibiotic discovery.
• Current methods struggle to rapidly identify effective compounds against emerging pathogens and resistant strains.

Purpose of the Study:

• To introduce ApexOracle, an AI model for predicting antibacterial activity and designing de novo antimicrobial molecules.
• To address limitations in identifying effective treatments for novel and drug-resistant bacterial infections.

Main Methods:

• Developed ApexOracle, an AI model integrating molecular features with pathogen-specific genomic and literature data.
• Utilized a foundational discrete diffusion language model and a dual-embedding framework for strain representation.
• Evaluated model performance in predicting antibacterial potency and designing novel compounds across diverse bacterial species.

Main Results:

• ApexOracle outperformed state-of-the-art methods in predicting antibacterial activity.
• Demonstrated reliable transferability to novel pathogens with limited antimicrobial data.
• Successfully generated de novo molecules with high predicted efficacy against priority bacterial threats.

Conclusions:

• ApexOracle offers a scalable strategy for combating antimicrobial resistance.
• The AI model accelerates the discovery of effective antibiotics against current and future infectious disease threats.