Tratopedia
繁中
Settings

Text size

Language

Theme

High contrast

Version

v1.230.2

The release this page was built from. It is what the service worker caches under.

Slide deck

Back

The Rise of Biomedical Foundation Models — 220 Models Surveyed

A survey of 220 biomedical foundation models built over four years finds 53.6% combine two or more data modalities, with natural language emerging as the connective interface. The bottleneck is scarce paired data and no standard multimodal benchmark, not a shortage of ideas.

  1. Research Brief · Nature Biotechnology · April 2026

    Biomedical Foundation Models Are Moving From One Modality to Many

    A survey of 220 biomedical foundation models

    Biomedical foundation models1 / 17

  2. What Happened

    What Counts as a Biomedical Foundation Model

    • Large-scale machine learning models trained on varied biomedical and clinical data.
    • That data spans genetic sequences, molecular profiling, biomedical imaging and electronic health records.
    • They learn generalisable representations to support downstream discovery and clinical work.
    • The survey covers four domains: natural language, imaging and signals, omics, and molecular sequences.
    • General-purpose LLMs are excluded; the focus is 220 biomedicine-specific models.

    Biomedical foundation models2 / 17

  3. What Happened

    220 models, four fields, more than half multimodal

    • 220models curated, over four years
    • 53.6%are multimodal — 118 combine two or more modalities
    • 17distinct data modalities in total
    • 4domains: language, imaging, omics, sequences

    The key observation

    Natural language is the modality most often paired with others, and is becoming the connective interface between data types. This runs through the whole paper: the emerging citation leaders and the underdeveloped directions both point back to the same structure.

    Biomedical foundation models3 / 17

  4. What Happened

    Confirmed

    StandingWhat happenedDetail
    ConfirmedThe survey is publishedChang, Cheng, Modi, Wang, Xu and Ma, in Nature Biotechnology — online 30 April 2026, in print that August at 44(8):1263–1266. DOI 10.1038/s41587-026-03135-y.
    Confirmed220 models, four fields, more than half multimodalEvery count and share on this page — the 220, the 118 (53.6%), the architecture split and the evaluation split — is stated in the paper's own running text and was read there.

    Biomedical foundation models4 / 17

  5. What Happened

    Confirmed, not verifiable here

    StandingWhat happenedDetail
    Confirmed, not verifiable hereWhich 220 models they areThe roster is Supplementary Table 1. The repository serves that file behind a challenge this project could not satisfy, so the list was never read model by model — the count is the paper's, not a recount.
    Confirmed, not verifiable hereThe citation countsThey come from Semantic Scholar and are printed only inside the paper's figure, which does not say when the snapshot was taken. They were read off that figure. Citation counts move, so treat them as one undated moment rather than a current ranking.

    Biomedical foundation models5 / 17

  6. Timeline

    From BioBERT to Evo 2

    Every model the survey names, dated to its own first publication

    1. 15 Feb 2020BioBERT — BERT re-trained on PubMed abstracts and PMC full text — is published; it becomes the survey's most-cited natural-language model.
    2. 13 Apr 2021ESM-1b shows that a language model trained on 250 million protein sequences, with no labels at all, learns real structural information; it is the ancestor of the survey's molecular-sequence leader.
    3. 26 Feb 2024scGPT, pretrained on more than 33 million individual cells, is published; it becomes the survey's omics leader.
    4. 19 Mar 2024UNI, pretrained on tissue images from over 100,000 whole-slide pathology scans, is published; it becomes the survey's imaging-and-signals leader.

    Biomedical foundation models6 / 17

  7. Timeline

    From BioBERT to Evo 2

    1. 16 Jan 2025ESM3 — a generative model reasoning jointly over protein sequence, structure and function — is published; the survey later names it a post-2025 riser.
    2. 4 Mar 2026Evo 2, trained on 9 trillion DNA base pairs spanning every domain of life, is published; the survey names it a second post-2025 riser.
    3. 30 Apr 2026Chang, Cheng, Modi, Wang, Xu and Ma publish the survey itself, online, in Nature Biotechnology.
    4. Aug 2026The survey appears in Nature Biotechnology's print issue, volume 44, number 8, pages 1263–1266.

    Biomedical foundation models7 / 17

  8. The Argument

    Encoders Dominate

    ArchitectureCountShare
    Encoder-based11451.8%
    Mixed6730.5%
    Decoder-based3114.1%
    Non-Transformer (e.g. Mamba)—3.6%

    Decoders are scarce not out of architectural preference but because of data: text-generation tasks need high-quality paired image–text annotation, and that is hard to obtain in biomedicine.

    Biomedical foundation models8 / 17

  9. The Argument

    Evaluation Is Still Narrow

    TaskShare
    Classification50.9%
    Report generation10.9%
    Question answering8.6%
    Segmentation7.3%

    The evaluation gap

    There are no multitask benchmarks, little human-in-the-loop assessment, and no combined treatment of data quality, interpretability and translational relevance.

    Biomedical foundation models9 / 17

  10. The Argument

    Citation Leaders by Domain

    * marks a multimodal model · counts are a Semantic Scholar snapshot the paper does not date

    DomainLeading models and citations
    Natural languageBioBERT 6,925 · MultiMedQA 3,707 · PubMedBERT 2,278
    Imaging and signalsUNI 1,141 · ConVIRT* 1,003 · MedCLIP* 794
    OmicsscGPT 846 · Geneformer 836 · scBERT 509
    Molecular sequencesESM-1b 2,807 · ProtTrans 1,207 · DNABERT 1,051

    Biomedical foundation models10 / 17

  11. The Argument

    The Risers Since 2025

    DomainRisers since 2025, and citations
    Natural languageMedGemma-27B 165 · MediPhi-Instruct 13
    Imaging and signalsBiomedCLIP* 498 · Quilt-1M* 216 · MedGemma* 165
    OmicsNicheformer* 81 · OmiCLIP* 58 · scGPT-spatial 34
    Molecular sequencesESM3 201 · Evo 2 197 · Borzoi 190

    Citation growth is being driven by models that integrate heterogeneous modalities rather than by specialists in a single data type. Five of the eleven risers are multimodal, and natural language is the only field with fewer than three of them.

    Biomedical foundation models11 / 17

  12. What Others Add

    What the Leading Models Are Actually Doing

    Independent papers behind the domain leaders

    • 250MProtein sequences ESM-1b learned from, with no labels at all (Rives et al., 2021).
    • 100k+Whole-slide pathology scans behind UNI's pretraining, across 20 tissue types (Chen et al., 2024).
    • 33MIndividual cells behind scGPT's pretraining (Cui et al., 2024).
    • 9TDNA base pairs behind Evo 2, spanning every domain of life (Brixi et al., 2026).

    Biomedical foundation models12 / 17

  13. What Others Add

    The Integrations Nobody Has Built Yet

    • Mechanistic interpretability

      Language ↔ sequence and omics

      • A language interface could assist scRNA-seq annotation, pathway enrichment and spatial transcriptomics.
    • Therapeutic targets

      Imaging ↔ omics

      • Integration of H&E pathology with omics remains weak.
      • The potential to link tissue morphology to molecular mechanism is considerable.
    • Regulatory layers

      DNA, RNA, protein, epigenome

      • The layers are only loosely connected, protein to the others most of all.

    Biomedical foundation models13 / 17

  14. What Others Add

    The bottleneck is not ideas

    The bottleneck is not ideas

    What blocks this is not a shortage of concepts but the state of the data: paired datasets are scarce, preprocessing and annotation are not standardised, and there is no standard multimodal benchmark. It is the same cause that explains the scarcity of decoders earlier.

    Biomedical foundation models14 / 17

  15. Conclusion

    What Gets a Model Adopted

    • It addresses a high-impact application: clinical language processing, protein modelling, digital pathology.
    • It is rigorously evaluated: testing across varied downstream tasks builds trust and reuse.
    • It is accessible: open pretrained weights, open code, a usable interface.
    • And newly decisive: it fits into agent-based AI workflows.

    Biomedical foundation models15 / 17

  16. Conclusion

    Still unsolved

    Still unsolved

    Data quality, cohort diversity, batch effects, privacy, hallucination and adversarial robustness all still limit real-world deployment. The authors recommend cross-disciplinary collaboration — biologists, clinicians and model developers together — and efficient adaptation of strong pretrained backbones in place of training from scratch.

    Biomedical foundation models16 / 17

  17. Conclusion

    What to remember

    A survey of 220 biomedical foundation models built over four years finds 53.6% combine two or more data modalities, with natural language emerging as the connective interface. The bottleneck is scarce paired data and no standard multimodal benchmark, not a shortage of ideas.

    Biomedical foundation models17 / 17