Protocol: random forest classification of microbiome profiles

Tier: D (cMD paper) · Type: atomic · Category: Machine Learning

What

Train and apply a random forest classifier to microbial abundance profiles, with the hyperparameters and feature-handling choices that the microbiome literature has settled on — stated as defaults with their provenance, not left to library defaults.

Why it matters

Split from the LODO protocol, per review. Random forests get used with ordinary cross-validation all the time; tying them to one validation scheme would have made both halves less useful. This protocol says how to fit the model; a validation protocol says how to evaluate it.

The value here is specificity to the domain. A generic "use a random forest" instruction helps nobody — the useful content is the parameter set that microbiome studies actually use, the fact that different published analyses use different ones, and what the differences do.

Defaults: follow Pasolli 2016

Two published parameterizations exist and they disagree. Adopt Pasolli's as the protocol's defaults.

	Pasolli et al. 2016 — the default	Manghi et al. 2025 (cMD) — variant
Trees	500	10,000
Features per split	√(total features)	1% of features
Impurity criterion	Gini	Entropy

Record the cMD variant in ## Notes as a published alternative with its source, so anyone reproducing
that paper knows what to change — but the protocol's stated defaults are Pasolli's, matching the
parameterization the citation/publication_doi pair points at.

Pasolli also specifies an embedded feature-selection strategy: rank features by importance, retrain on
progressively larger subsets of {5, 10, 20, … 200}, and keep the subset size that maximizes accuracy.
Include it as a defined optional step with its own rationale, not folded silently into the fit.

Source material

• Pasolli, Truong, Malik, Waldron & Segata (2016), PLoS Comput Biol, 10.1371/journal.pcbi.1004977 — the microbiome parameterization and the feature-selection strategy
• waldronlab/curatedMetagenomicDataAnalyses — cMD3_paper_analyses/ML/ml_tests_on_diseases_rf.py, figure4_complete_ml.py; the external metaml tool
• Applied in: 10.1038/s41467-025-66888-1

Scope

In: the hyperparameters as stated defaults with their source; which features enter the model and on what scale (and therefore whether a transformation protocol is required or optional); class imbalance handling; the embedded feature-selection procedure, as an optional step with its own rationale; what the model outputs and what feature importances may and may not be read as.

Out: the validation scheme — LODO or otherwise; performance metrics.

Frontmatter starting point

type: "atomic"
category: "Machine Learning"
citation: "10.1023/A:1010933404324"        # Breiman 2001 — random forests. VERIFY
publication_doi: "10.1371/journal.pcbi.1004977"   # Pasolli 2016 — source of the microbiome parameterization
protocols_used:
  - name: "prevalence-filtering"
  - name: "clr-transformation"      # optional: see note
tags: [machine-learning, random-forest, classification, microbiome]

Breiman proposed random forests; Pasolli proposed this parameterization for microbiome data. The
one-citation rule puts Breiman in citation and Pasolli in publication_doi — this split is confirmed
as the intended reading. The same pattern — a general method with a domain-specific published
parameterization — will recur, so it is worth writing up in ## Notes as a precedent.

Acceptance criteria

• Every hyperparameter is a stated default with a named source
• Pasolli's parameterization is the stated default; the cMD variant is recorded in ## Notes with its source
• The embedded feature-selection strategy is a defined optional step, not implicit
• Contains no validation scheme — it composes with lodo-cross-validation, it does not assume it
• States plainly what random forest feature importance does not establish

---

Cite the method's origin, not its users

PROTOCOL_STANDARD.md is explicit: an atomic protocol carries "strictly 1 citation... corresponding
to the primary literature where the method was originally published." Find the paper that proposed
the method. Do not cite a paper that merely applied it — including the BugSigDB and curatedMetagenomicData
papers, which are the source of the analysis these protocols were extracted from but almost never the
source of the method.

Tracing a method back to its first publication is real work, and it is part of the task, not a
formality. Three things to expect:

• Candidate DOIs in this issue are leads, not answers. Anything marked VERIFY has not been checked.
• Some methods predate modern citation practice or have no single identifiable origin. If that is
  genuinely the case, say so in the pull request rather than reaching for a convenient recent paper.
  Raise it as an issue in waldronlab/agent-protocol-standard — the standard may need a way to express
  "classical method, no primary source".
• If you cannot name one paper that proposed everything the protocol does, it is more than one
  protocol. That test has now split four protocols out of this batch: enrichment into three methods,
  filtering from transformation, LODO from random forest, and PERMANOVA from ANOSIM.

Where the lab's own paper genuinely did propose the method — the oral-to-gut score, and LODO
cross-validation in Pasolli et al. 2016 — citing it is correct. That is the exception, not the pattern.

Before you start

Read CONTRIBUTING.md and
PROTOCOL_STANDARD.md.
The format is defined in the standard repo, not this one. Protocols are prose, not code: they say what to do
and why, precisely enough that two people — or two agents, in two languages — get the same answer. The existing
independent-filtering-variance
protocol is the model to imitate for tone and level of detail.

Validate locally before opening the PR:

git clone https://github.com/waldronlab/agent-protocol-standard.git
Rscript agent-protocol-standard/scripts/validate-protocol.R protocols

Read CONTRIBUTING.md and PROTOCOL_STANDARD.md first, then inspect the independent-filtering-variance protocol for style. Review cMD3_paper_analyses/ML/ml_tests_on_diseases_rf.py and figure4_complete_ml.py alongside the cited Pasolli source, and run the documented R validation command. Done means the protocol meets every acceptance criterion and passes local validation.