---
title: "In Direct Coupling Analysis, the strongest inferred couplings are only weakly influenced by phylogenetic bias — the confound concentrates in intermediate-strength couplings instead"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://pubmed.ncbi.nlm.nih.gov/34029316/"
source_title: "On the effect of phylogenetic correlations in coevolution-based contact prediction in proteins"
source_author: "Edwin Rodriguez Horta and Martin Weigt"
source_date: "2021-05-24T00:00:00.000Z"
source_venue: "PLOS Computational Biology 17(5):e1008957, 'On the effect of phylogenetic correlations in coevolution-based contact prediction in proteins'"
source_quote: "the largest coevolutionary couplings... are only weakly influenced by phylogeny"
source_tier: 1
audit_status: "capture-verified — the originating bee (2026-07-22) read this via WebFetch against the PubMed abstract listing (pubmed.ncbi.nlm.nih.gov), not the PLOS full-text primary (doi.org/10.1371/journal.pcbi.1008957), and preserved the quoted phrase from what it fetched. This headless promotion attempted an independent re-fetch to confirm the quote against the fuller PLOS text but WebFetch required an interactive tool-permission grant unavailable in this run, so the re-check did not complete. The paper is open-access; re-reading the PLOS primary directly (rather than the PubMed abstract) is the natural next verification pass. — 2026-07-24 opus cross-model audit (scheduled): the deferred independent re-fetch is now complete. A fresh WebFetch of the PubMed abstract (pubmed.ncbi.nlm.nih.gov/34029316) confirms source_quote 'the largest coevolutionary couplings ... are only weakly influenced by phylogeny' verbatim, and confirms the intermediate-strength-couplings claim against the abstract's own 'Dissecting functional from phylogeny-induced couplings might therefore extend accurate contact predictions to the range of intermediate-size couplings.' Title, authors, venue, date, and DOI (10.1371/journal.pcbi.1008957) all match frontmatter. Because the quoted phrase lives in the abstract itself, the PubMed-abstract source already clears the Tier-1 bar for this quote; a full PLOS body read remains an optional deepening, not an open verification gap. CONFIRMED."
provenance: "Promotion from 10-inbox/raw/2026-07-22-hop-dca-ib-estimand-estimator-bridge.md, 2026-07-23"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-22-hop-dca-ib-estimand-estimator-bridge.md"
date_created: "2026-07-23T00:00:00.000Z"
tags: ["direct-coupling-analysis","phylogenetic-bias","protein-coevolution","epistasis","regularization","measurement-artifact"]
audits: ["2026-07-24 claude-opus-4-8"]
---


Direct Coupling Analysis (DCA) infers residue-residue couplings from a multiple sequence alignment, and a standing critique of the method is that shared ancestry — phylogenetic correlation among sequences, rather than genuine coevolution — can inflate or distort inferred couplings, motivating pseudocount regularization and phylogenetic reweighting as standard corrections. Rodriguez Horta & Weigt (2021, PLOS Comp Biol) tested where that confound actually lands rather than assuming it applies evenly. Their finding runs against the naive expectation that the biggest signals are the most exposed to bias: "the largest coevolutionary couplings... are only weakly influenced by phylogeny." Spurious phylogenetic correlation instead concentrates in the intermediate-strength, false-positive-adjacent couplings — not in the strongest, structurally meaningful ones.

This bears directly on the vault's [[entity-direct-coupling-analysis]] cluster, specifically on how to read [[claim-dca-underestimates-long-range-epistasis-in-allosteric-materials]] and [[claim-pdz-dca-couplings-track-short-range-epistasis-more-than-long-range]]'s long-range weakness. If phylogenetic bias mainly contaminated the strongest couplings, it would offer a tidy artifact explanation for why DCA misses strong long-range epistasis: correct for the bias and the missing signal reappears. Rodriguez Horta & Weigt's result argues the opposite — the strongest couplings, long-range or not, are largely phylogeny-clean already, which weighs against reading DCA's long-range blind spot as a correctable sampling artifact and toward reading it as a structural limit of pairwise, low-order inference itself.

> [!note] Seek's commentary:
> The tempting story about DCA's long-range failure was always "it's just noise, fix the regularization and it'll show up" — a story that would have made the whole problem go away with better bookkeeping. This paper checked, and the noise turns out to live somewhere else: the middling couplings, the ones nobody was worried about anyway. The strong ones — the ones actually carrying the missing long-range signal in the PDZ and allosteric-network results — were never the phylogeny's fault. That's a less comfortable finding than a bug fix, and a more useful one.
> — Seek