{
  "abstract": "Membrane protein interactions underpin essential cellular processes, yet the residues mediating these interactions remain difficult to identify at scale due to the paucity of experimentally resolved protein complexes within hydrophobic bilayers. Here, we present CHORUS, a sequence-only framework for predicting interaction interfaces in membrane proteins without requiring structural coordinates, multiple sequence alignments, or known interacting partners. By synthesizing complementary latent representations from three protein language models, CHORUS outperforms 18 established sequence- and structure-based methods by a chain-macro average precision value of 0.643 versus 0.613 for ScanNet-MSA on an independent, homology-restricted test set of 4,655 protein chain contexts, with a particularly pronounced advantage within the membrane-embedded interior (0.746 versus 0.645). Using six mammalian mitochondrial respiratory supercomplexes with >100 independent chains, we show that CHORUS-identified intersubunit interaction interfaces accurately delineate the boundaries between buried protein-protein contact surfaces and surrounding lipid-facing regions in complex molecular systems (0.945-0.984). We subsequently applied CHORUS at proteome scale to chart a global landscape of membrane-protein interaction interfaces for 80,619 expert-reviewed membrane protein entries, 70,247 unique sequences, 4,918 species, and 33.7 million residues across the tree of life. The resulting open-access CHORUS Atlas (https://chorus.atlas.sunnylab.org) provides predicted residue interactions across the membrane proteome. Together, CHORUS establishes a sequence-based foundation for systematic investigation of membrane protein assembly and function, potentially facilitating interactome-scale knowledge discovery and allosteric therapeutic design.",
  "abstract_shown": true,
  "affiliations": [
    "Hong Kong University of Science and Technology (Guangzhou)"
  ],
  "article_type": "New results",
  "arxiv": null,
  "author_total": 3,
  "authors": [
    {
      "name": "J. Zhao"
    },
    {
      "name": "S. Huang"
    },
    {
      "affs": [
        0
      ],
      "name": "J. Sun"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/e11d9j4e0071a.bib",
  "cited_by": 0,
  "doi": "10.64898/2026.10.04.756538",
  "fulltext_url": "https://www.biorxiv.org/content/10.64898/2026.10.04.756538v1.full",
  "id": "e11d9j4e0071a",
  "issns": [],
  "landing_url": "https://www.biorxiv.org/content/10.64898/2026.10.04.756538v1",
  "language": null,
  "licence": "CC-BY-NC-ND",
  "licence_label": "CC BY-NC-ND",
  "licence_source": "article",
  "licence_url": "https://creativecommons.org/licenses/by-nc-nd/4.0/",
  "metadata_licence": null,
  "native_subjects": [
    [
      "biorxiv",
      "bioinformatics"
    ],
    [
      "openalex",
      "1312|Machine Learning in Bioinformatics"
    ]
  ],
  "openalex": "W7221465435",
  "pdf_url": "https://www.biorxiv.org/content/10.64898/2026.10.04.756538v1.full.pdf",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-04",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/e11d9j4e0071a.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:27:12Z",
      "native_id": "10.64898/2026.10.04.756538",
      "source": "biorxiv",
      "url": "https://www.biorxiv.org/content/10.64898/2026.10.04.756538v2"
    }
  ],
  "subjects": [
    {
      "field": "biology",
      "origin": "biorxiv:bioinformatics",
      "subfield": "bioinformatics"
    }
  ],
  "title": "CHORUS: A proteome-scale atlas of membrane interaction interfaces constructed using cross-protein language model embeddings from sequences alone",
  "title_text": "CHORUS: A proteome-scale atlas of membrane interaction interfaces constructed using cross-protein language model embeddings from sequences alone",
  "type": "preprint",
  "updated_at": "2026-11-04T00:21:18Z",
  "url": "https://research.draconitic.com/paper/e11d9j4e0071a",
  "venue": "bioRxiv",
  "versions": []
}