Practically Significant Method Comparison Protocols for Machine Learning in Small Molecule Drug Discovery.

Area of Science:

• Computational chemistry
• cheminformatics
• machine learning
• drug discovery

Background:

• Machine learning (ML) models predict molecular properties, aiding drug discovery by replacing experiments.
• Current ML method comparisons lack standardization, hindering reproducibility and adoption.
• Robust evaluation is crucial for high-stakes decisions in small molecule drug discovery.

Purpose of the Study:

• To propose guidelines for rigorous and domain-appropriate comparison of ML methods for small molecule property modeling.
• To promote the development and adoption of reliable ML tools in drug discovery.

Main Methods:

• Development of a set of guidelines for ML method comparison.
• Inclusion of annotated examples using open-source software tools.
• Focus on statistically rigorous protocols and domain-appropriate performance metrics.

Main Results:

• A foundational framework for robust ML benchmarking in small molecule property prediction.
• Guidelines designed to incentivize rigorous techniques and ensure replicability.
• Open-source examples facilitating practical implementation.

Conclusions:

• Standardized guidelines are essential for advancing ML in small molecule drug discovery.
• Rigorous benchmarking ensures the development of impactful and reliable in silico tools.
• Adoption of these guidelines will foster trust and accelerate innovation in the field.