Transfer learning predicts species-specific drug interactions in emerging pathogens.

Area of Science:

• Microbiology
• Computational Biology
• Pharmacology

Background:

• Machine learning (ML) is crucial for identifying drug combinations against resistant bacteria.
• Existing ML models struggle with pathogens lacking sufficient training data.

Purpose of the Study:

• To develop a novel framework (TACTIC) for predicting drug interactions in under-studied bacteria using transfer learning and crowdsourcing.
• To identify novel synergistic drug combinations effective against Gram-negative and non-tuberculous mycobacteria (NTM) pathogens.

Main Methods:

• Developed the TACTIC framework integrating transfer learning and crowdsourcing on 2,965 drug interactions across 12 bacterial strains.
• Applied TACTIC to predict ~600,000 drug interactions across diverse bacterial species and metabolic environments.
• Experimentally validated predicted synergistic combinations against *M. abscessus*.

Main Results:

• TACTIC outperformed traditional ML models in predicting drug interactions for species with limited data.
• Identified selective synergistic drug combinations against Gram-negative pathogens like *A. baumannii* and NTM.
• Validated synergistic combinations including clarithromycin, ampicillin, and mecillinam against *M. abscessus*.

Conclusions:

• TACTIC enables effective drug combination prediction for emerging pathogens with sparse data.
• Identified promising drug combinations for combating antibiotic resistance in Gram-negative and NTM infections.
• Proposed novel synergistic combinations for treating bacterial eye infections (endophthalmitis).