CODARFE

CODARFE applies compositional data analysis with recursive feature elimination to select sparse microbial predictors and predict continuous environmental variables from taxonomic microbiome compositions.

Key Features:

• Compositional data analysis: Operates on compositional microbiome data and taxonomic compositions.
• Recursive feature elimination: Implements recursive feature elimination to produce sparse sets of microbial taxa predictive of environmental factors.
• Prediction of continuous environmental factors: Trains models to predict continuous environmental variables from microbiome samples.
• Comparative performance: Outperformed four state-of-the-art tools in predictor selection across 21 of 24 databases based on correlation metrics and recovered at least 7% more previously recognized bacteria linked to environmental factors in human data.
• Cross-study validation: Performs cross-study experiments (e.g., ginseng field vs. cattle for arable soil; HIV vs. Crohn's disease for the human gut) with a mean absolute percentage error of 11%.

Scientific Applications:

• Microbiome–environment association analysis: Identifying microbial taxa correlated with continuous environmental gradients in ecology, agriculture, and human health.
• Environmental variable prediction: Predicting continuous environmental variables from new microbiome samples.
• Cross-study biomarker validation: Validating candidate microbial biomarkers across studies and sample types, as demonstrated for arable soil and human gut comparisons.

Methodology:

Applies recursive feature elimination on compositional taxonomic microbiome data to select sparse predictors, trains predictive models for continuous environmental variables, evaluates performance using correlation metrics and mean absolute percentage error, and performs cross-study train/test validations.