chemmodlab: a cheminformatics modeling laboratory R package for fitting and assessing machine learning models.

Area of Science:

• Cheminformatics
• Machine Learning
• Statistical Modeling

Background:

• Comparing machine learning models is crucial for research.
• Existing pipelines can be complex and time-consuming.
• Standardized assessment methods are needed in cheminformatics.

Purpose of the Study:

• To introduce chemmodlab, an R package for streamlined machine learning model fitting and assessment.
• To facilitate easy comparison of model utility for researchers.
• To provide novel visualization tools for performance differences.

Main Methods:

• Implementation of expert-accepted model fitting and assessment methods.
• Development of assessment utilities including accumulation curves and performance measures.
• Introduction of a multiple comparisons similarity plot for visualizing significant differences.

Main Results:

• chemmodlab streamlines the model comparison pipeline.
• The package includes functions for accumulation curves and performance metrics.
• A novel plot visualizes statistically significant performance differences using repeated k-fold cross-validation.

Conclusions:

• chemmodlab enhances the ability of researchers to compare machine learning models.
• The package offers broad utility beyond cheminformatics.
• It provides efficient and statistically sound methods for model assessment.