Focusing on Data to Improve Machine Learning-Guided Antibiotic Discovery.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Machine learning (ML) shows promise for accelerating the discovery of novel antibiotic compounds.
• Current algorithmic progress has led to limited real-world performance improvements in antibiotic discovery.
• Significant advancements are needed beyond algorithms for effective antibiotic development.

Purpose of the Study:

• To review data-centric strategies for building effective machine learning pipelines in antibiotic discovery.
• To highlight the importance of data acquisition, representation, and expert interpretation.
• To guide the development of robust, reliable, and biologically grounded ML models for identifying new antibiotics.

Main Methods:

• Focus on data acquisition and quality control for ML models.
• Emphasis on data representation techniques relevant to chemical compounds.
• Integration of domain expert knowledge for model output interpretation.
• Discussion of standardized data curation, benchmarking, and publication practices.

Main Results:

• Algorithmic advances alone offer modest gains; data-centric approaches are crucial.
• Improved data quality and representation enhance ML model performance in drug discovery.
• Expert interpretation of ML outputs is vital for biological relevance and clinical application.
• Standardization in data and practices is essential for field-wide progress.

Conclusions:

• Data-centric choices are paramount for developing successful ML pipelines for antibiotic discovery.
• Improving data acquisition, representation, and expert interpretation will drive greater gains.
• Standardized practices are necessary to unlock the full potential of ML in addressing the antibiotic resistance crisis.