Overview

This tutorial walks through a complete experiment using the Franzosa et al. 2019 inflammatory bowel disease dataset, a gut microbiome cohort with three clinical groups: Crohn’s disease (CD), ulcerative colitis (UC), and non-IBD controls. Steps that are dataset-specific are noted where relevant.

Steps

Step	Description
Data preparation	Download the example cohort files and convert them into the profile format accepted by mllabiome
Experiment configuration	Define the classification target, preprocessing pipeline, model search space, and cross-validation strategy
Running the experiment	Execute the experiment script and monitor progress
Experiment output structure	Understand the directory layout and every output artefact produced by the sweep
Resuming and extending an experiment	Pick up an interrupted run or expand an existing sweep with new configurations
Post-evaluation	Load evaluation results and rank configurations by generalisation performance
Post-explain: global	Compute SHAP-based global feature importance across the best configurations
Post-explain: local	Generate per-sample explanations for individual predictions
Ensemble sweep	Search for an ensemble of top-performing models that improves on any single model