Machine learning to design antimicrobial combination therapies: Promises and pitfalls

Area of Science:

• Pharmacology
• Computational Biology
• Infectious Diseases

Background:

• Antimicrobial resistance (AMR) necessitates novel therapeutic strategies.
• Drug repurposing and combination therapies offer potential solutions.
• The vast number of potential drug combinations poses a significant challenge.

Purpose of the Study:

• To compare machine learning (ML) approaches for designing drug combination therapies.
• To identify optimal ML strategies based on input data types (drug properties, microbial response, infection microenvironment).
• To compile publicly available drug interaction datasets for AMR research.

Main Methods:

• Comparative analysis of ML algorithms for drug combination prediction.
• Evaluation of ML models based on input data modalities.
• Literature review and compilation of relevant AMR drug interaction datasets.

Main Results:

• Different ML approaches show varying efficacy depending on the input data.
• Drug properties, microbial response, and microenvironment data each offer unique advantages for ML models.
• A comprehensive list of AMR-relevant drug interaction datasets is provided.

Conclusions:

• ML holds significant promise for identifying synergistic drug combinations against AMR.
• Future strategies should incorporate in vivo conditions, sequential dosing, and deep learning for enhanced efficacy and interpretability.
• Addressing limitations in current ML methods is crucial for developing next-generation combination therapies.