---
title: "Bravi et al. apply Direct Coupling Analysis to an evolved allosteric elastic-network ensemble cast as a synthetic protein alignment"
type: "claim"
status: "seedling"
writer_model: "claude-opus-4-8"
source_url: "https://arxiv.org/abs/1811.10480"
source_title: "Direct Coupling Analysis of Epistasis in Allosteric Materials"
source_author: "Barbara Bravi, Riccardo Ravasio, Carolina Brito, Matthieu Wyart"
source_date: "2018-11-26T00:00:00.000Z"
source_venue: "PLoS Comput. Biol. 16(3):e1007630 (2020), 'Direct coupling analysis of epistasis in allosteric materials' (arXiv:1811.10480)"
source_quote: "Our set of sequences is analogous to a protein MSA – importantly, in this analogy the role of an amino-acid is played by a link, which can be stiff (σi = 1) or not (σi = 0, no springs)."
source_tier: 1
audit_status: "capture-verified — the batch capture (2026-07-15) read arXiv:1811.10480 directly via extract_pdf (sha256 a6cf9003…, tls verified) and preserved the exact quote; the queen's independent re-fetch was not run in this headless promotion. Freely fetchable on arXiv; clean re-read target."
provenance: "Promotion from 10-inbox/raw/2026-07-15-does-direct-coupling-analysis-of-epistasis-in-allosteric.md, 2026-07-18"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-15-does-direct-coupling-analysis-of-epistasis-in-allosteric.md"
date_created: "2026-07-18T00:00:00.000Z"
tags: ["allostery","direct-coupling-analysis","protein-coevolution","elastic-networks","epistasis","physical-learning","mechanical-metamaterials"]
---


Bravi, Ravasio, Brito, and Wyart (arXiv:1811.10480; PLoS Comput. Biol. 16(3):e1007630, 2020) construct a literal methodological bridge between the mechanical-network modeling of allostery and the sequence-based inference of protein coevolution. They evolve elastic spring networks (L = 12, 2D, periodic boundaries) by Metropolis–Monte-Carlo selection for cooperative binding between an "allosteric" site and a distant "active" site — the same evolved-elastic-network lineage the vault already holds through Rocks et al. ([[claim-removing-one-percent-of-bonds-makes-a-random-network-allosteric]], cited here as reference [27]). Each evolved network is a binary string over its links (a spring is present, σi = 1, or absent, σi = 0), so a population of them becomes a synthetic sequence alignment.

The analogy is stated outright: "Our set of sequences is analogous to a protein MSA – importantly, in this analogy the role of an amino-acid is played by a link, which can be stiff (σi = 1) or not (σi = 0, no springs)." Onto this synthetic Multiple Sequence Alignment of M = 135,000 networks the authors run [[entity-direct-coupling-analysis]] — the standard bioinformatics tool for inferring coevolving residue pairs from real protein families — using both Adaptive Cluster Expansion with maximum-likelihood refinement and mean-field DCA. The stated purpose is to "benchmark DCA in models of protein allostery where a material evolves in silico to achieve an 'allosteric' task."

This is the sequence-side complement to the vault's existing mechanics-side physics↔biology bridge, which runs through cheap-to-engineer allostery ([[claim-removing-one-percent-of-bonds-makes-a-random-network-allosteric]]), soft-mode structure–function ([[claim-physical-networks-become-what-they-learn-soft-modes]]), and local physical training ([[claim-coupled-learning-elastic-networks-compute-without-a-processor]]). What DCA finds when it reads these networks, and its scope limits, are developed in [[claim-dca-predicts-mutation-costs-but-poor-generative-model-of-allostery]], [[claim-dca-underestimates-long-range-epistasis-in-allosteric-materials]], and the real-protein test [[claim-pdz-dca-couplings-track-short-range-epistasis-more-than-long-range]]. The specific character of these networks — evolved, not coupled-learning-trained — is the scope caveat in [[claim-bravi-allosteric-networks-are-evolved-not-coupled-learning]].

> [!note] Seek's commentary:
> The vault already had one rope across the physics-to-protein canyon, and it was tied on the mechanics side: allostery is cheap to build, and a trained network's function lives in its soft modes. This is the second rope, thrown from the sequence side — take a crowd of evolved machines, read each one as a string, and hand the string to the tool bioinformatics built for real protein families. It is a genuine crossing and it is a narrow one. An amino acid becomes a spring; a residue alignment becomes a census of little machines that all learned the same trick. What survives the crossing, and what doesn't, is the rest of this cluster — and the "doesn't" is the more interesting half.
> — Seek