{
  "abstract": "Single-cell foundation models (scFMs) are commonly benchmarked using final-layer embeddings, although useful depth may depend on the task. We evaluate Geneformer, scBERT, scGPT, and scFoundation as frozen feature extractors for cell-type and perturbation-response prediction using layer-wise probes, leave-one-dataset-out (LODO) selection, readout controls, representation diagnostics, centered kernel alignment (CKA), and retrieval/collapse analyses. Intermediate layers significantly improve cell-type accuracy for scGPT and scBERT in several settings, while Geneformer and scFoundation generally show performance comparable to their final layers. For perturbation prediction, LODO identifies layer 1 as an earliest-on-par candidate across all four models, without establishing universal superiority. Paired scGPT CLS and mean-pooling analyses show that readout affects preferred depth. CKA and representation metrics reveal model- and task-dependent geometric changes, while retrieval and collapse checks support the robustness of perturbation findings. We recommend reporting extraction depth, readout, selection protocol, and task-specific diagnostics. A validated non-final layer may also reduce computation, although efficiency was not directly benchmarked.",
  "abstract_shown": true,
  "affiliations": [
    "Department of Computational Medicine and Bioinformatics, University of Michigan"
  ],
  "article_type": "New results",
  "arxiv": null,
  "author_total": 3,
  "authors": [
    {
      "affs": [
        0
      ],
      "name": "M.-C. C. Wen"
    },
    {
      "name": "Y. Lyu"
    },
    {
      "name": "A. Robbins"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/9n6ezm9x3j5sq.bib",
  "cited_by": 0,
  "doi": "10.64898/2026.10.05.756880",
  "fulltext_url": "https://www.biorxiv.org/content/10.64898/2026.10.05.756880v1.full",
  "id": "9n6ezm9x3j5sq",
  "issns": [],
  "landing_url": "https://www.biorxiv.org/content/10.64898/2026.10.05.756880v1",
  "language": null,
  "licence": "CC-BY",
  "licence_label": "CC BY",
  "licence_source": "article",
  "licence_url": "https://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": null,
  "native_subjects": [
    [
      "biorxiv",
      "bioinformatics"
    ],
    [
      "openalex",
      "1312|Single-cell and spatial transcriptomics"
    ]
  ],
  "openalex": "W7221540172",
  "pdf_url": "https://www.biorxiv.org/content/10.64898/2026.10.05.756880v1.full.pdf",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-05",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/9n6ezm9x3j5sq.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:27:17Z",
      "native_id": "10.64898/2026.10.05.756880",
      "source": "biorxiv",
      "url": "https://www.biorxiv.org/content/10.64898/2026.10.05.756880v1"
    }
  ],
  "subjects": [
    {
      "field": "biology",
      "origin": "biorxiv:bioinformatics",
      "subfield": "bioinformatics"
    }
  ],
  "title": "Beyond the Final Layer: Evaluating Single-Cell Foundation Model Representations for Molecular Prediction",
  "title_text": "Beyond the Final Layer: Evaluating Single-Cell Foundation Model Representations for Molecular Prediction",
  "type": "preprint",
  "updated_at": "2026-10-30T12:27:17Z",
  "url": "https://research.draconitic.com/paper/9n6ezm9x3j5sq",
  "venue": "bioRxiv",
  "versions": []
}