Machine learning for microbiome research
Build, compare, combine and explain predictive models from microbiome abundance data.
A research library for finding reproducible patterns in microbiome data and testing whether modelling choices generalize beyond the samples used to build them.
Does a new algorithm really improve on what already exists?
Place a new estimator beside established approaches and evaluate them under the same data, preprocessing rules and held-out samples. Model selection stays inside the training data, while final comparisons are made on observations that did not guide that selection.
Start with microbiome abundance profiles and study metadata
Samples from a study cohort become rows in an abundance matrix. mllabiome starts from these processed profiles; sequencing and taxonomic profiling happen upstream.
From candidate models to interpretable results
The detailed workflow stays below the introduction so the first view explains the purpose before the machinery.
Compare ways of representing and modelling the data
Different microbiome representations and learning algorithms are evaluated under the same validation design.
MPDR × MPMA-BCombine complementary models
Eligible base models can enter ensemble selection using evidence from the training side of the evaluation.
MPMA-EUnderstand which features drive predictions
Global explanations summarize patterns across held-out samples; local explanations show contributions for individual predictions.
ExplainabilityExplore, evaluate, explain and report
Research use only. Not a medical device. Study-specific models and biological interpretations require independent validation.