---
title: "The allosteric networks in Bravi et al.'s DCA study are evolved by Monte-Carlo optimization, not trained by coupled learning"
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: "Such networks are evolved by changing the position of springs according to a Metropolis-Monte-Carlo routine to maximize F."
source_tier: 1
audit_status: "capture-verified — the batch capture (2026-07-15) read arXiv:1811.10480 directly via extract_pdf (tls verified) and preserved the exact quote and the reference-list citation of Rocks et al. [27]; 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: ["physical-learning","coupled-learning","allostery","elastic-networks","evolutionary-optimization","scope-distinction"]
---


A scope distinction that matters for how Bravi, Ravasio, Brito, and Wyart's DCA result ([[claim-bravi-2020-applies-dca-to-evolved-allosteric-networks-as-synthetic-msa]]) connects to the vault's physical-learning cluster: the allosteric networks it studies are produced by an in-silico evolutionary search, not by a local physical learning rule. The paper's own description — "Such networks are evolved by changing the position of springs according to a Metropolis-Monte-Carlo routine to maximize F" — places the method in the evolved/designed elastic-network tradition of Yan, Ravasio, Brito & Wyart (2017, 2018) and of Rocks et al. (PNAS 2017), which the paper cites as reference [27] and which the vault already holds as [[claim-removing-one-percent-of-bonds-makes-a-random-network-allosteric]].

That is a different mechanism from coupled learning, the local contrastive rule in which each spring updates its own rest length using only its free-vs-clamped response, with no external optimizer and no global error signal ([[claim-coupled-learning-elastic-networks-compute-without-a-processor]]). Both produce networks that perform allostery, and both may land in the same soft-mode structural regime ([[claim-physical-networks-become-what-they-learn-soft-modes]]) — but a global fitness-maximizing search over spring configurations and a decentralized in-situ learning rule are not the same process, even when they reach a similar artifact.

The distinction guards against overstating the bridge. Read broadly — "physically-instantiated network models of allostery evolved or optimized for mechanical function" — this DCA study does bridge the vault's mechanical-allostery lineage to protein-sequence coevolution. Read in the narrower, technical sense the vault reserves for coupled learning, it does not test that specific class of network. Whether reaching the same structure by different routes implies anything shared beyond the structure is the standing question of [[backpropagation-gap]].

> [!note] Seek's commentary:
> This is the note I most wanted to write, because it is the one that keeps the headline from inflating. The topic question asked whether this paper bridges "physical-learning networks" to protein coevolution, and the tidy answer is yes — but tidy is doing work there. These networks were evolved by a Monte-Carlo routine cranking a fitness function, which is a designer with a search algorithm, not a material teaching itself by a local rule with nobody in the loop. The vault means something specific by physical learning, and this isn't quite it. Same destination — an allosteric spring network — reached by an outside optimizer rather than by physics doing its own [[entity-credit-assignment|credit assignment]]. Flatten that and you get a stronger claim than the paper supports; keep it and you get an honest one. I kept it.
> — Seek