Google DeepMind (AlphaFold, Gemini) Healthcare Breakthroughs: Every Medical and Drug Discovery Move Since 2016

Google DeepMind has one of the broadest health records of any frontier AI lab, from a troubled NHS app in 2016 to a Nobel Prize for AlphaFold. This page logs every move we could source and grades how far each result has been proven.

Presented By Our Partners

What Has Google DeepMind Done in Healthcare?

Google DeepMind has one of the broadest health and biology records among frontier labs, though Microsoft Research published biomedical work a decade earlier. Its record splits into two tracks: biology tools such as AlphaFold, which scientists use every day, and clinical AI such as Med-PaLM and AMIE, which has strong published results but little real-world patient testing so far.

This page treats Google DeepMind, Google Research and the former Google Health group as one story, because the same teams and models move between them. The 2020 mammography paper, for example, lists both Google Health and DeepMind authors, and the AMIE studies list both Google Research and Google DeepMind. Where a result came mainly from Google Research or Google Health, we say so. Gemini is the base model under newer systems: Google built its AI co-scientist on Gemini, and the 2026 AMIE disease management study used Gemini’s long context to ground answers in clinical guidelines (PMID 42310463).

The early years were rocky. DeepMind’s first health project, the Streams app built with the Royal Free London NHS trust, led the UK Information Commissioner’s Office to rule in July 2017 that the trust’s trial with DeepMind did not comply with data protection law. In November 2018 the Streams team moved into Google. The biology track went better: AlphaFold led the CASP13 blind assessment in 2018, its second version reached near experimental accuracy at CASP14 in 2020, and the AlphaFold database built with EMBL-EBI now covers more than 200 million predicted structures.

Google DeepMind Healthcare Timeline: Every Move Since 2016

Google DeepMind’s Biggest Results, and How Far Each Has Been Proven

AlphaFold is the only Google DeepMind health or biology result that has been independently confirmed and is in wide daily use. The clinical models are peer-reviewed, and two have now been studied in real patients, but none has been shown to improve patient outcomes in a randomized trial.

AlphaFold protein structure prediction. Independently replicated, Research tool. In the CASP13 blind assessment, the first AlphaFold built high-accuracy structures for 24 of 43 free modeling domains, the hardest category, against 14 for the next best method (PMID 31942072). AlphaFold 2 reached accuracy competitive with experimental structures in most CASP14 cases (PMID 34265844). An outside team that compared predictions with experimental crystal maps found many matched remarkably closely, but some high-confidence predictions were wrong in places, and advised treating them as strong hypotheses rather than replacements for experiments (PMID 38036854). What would prove more: a medicine whose design clearly depended on an AlphaFold structure reaching approval.

AMIE, the diagnostic conversation model. Tested in people, Clinical care. In a simulated study with patient actors, AMIE beat 20 primary care physicians on 30 of 32 axes rated by specialists, using text chat (PMID 40205050). In October 2026, a prospective feasibility study at Beth Israel Deaconess enrolled 114 patients, and 98 completed both an AMIE chat and their urgent care visit. Supervising physicians needed zero safety stops, noted one hallucination and added clinical information in five chats (PMID 42849491). This was a single-site, single-arm study funded by Alphabet, so it shows safety and feasibility, not better care. Separately, in a June 2026 Nature study with simulated patients, a Gemini-grounded AMIE was non-inferior to 21 primary care physicians on management reasoning across 100 multivisit cases (PMID 42310463). What would prove more: a randomized trial measuring diagnosis accuracy or outcomes.

Diabetic eye screening in Thailand (Google Health). Tested in people, Clinical care. Across 7,651 patients at nine primary care sites, the system found vision-threatening diabetic retinopathy with 91.4% sensitivity, against 84.8% for retina specialists who over-read the same images, with similar specificity (PMID 35272972). What would prove more: evidence that screening this way prevents vision loss.

Featured Partner

Invest in the Infrastructure Behind Modern Medicine

As healthcare expands beyond hospital walls, the buildings and campuses supporting that shift are generating compelling returns for investors who move early. The Healthcare Real Estate Fund offers qualified investors direct access to a curated portfolio of medical office, outpatient, and specialty care facilities.

Learn More →

Breast cancer screening and Moorfields eye referral. Peer-reviewed, Clinical care. On US and UK data, the 2020 mammography model cut false positives by an absolute 5.7% and 1.2% and false negatives by 9.4% and 2.7% (PMID 31894144). The 2018 Moorfields model matched or beat experts on referral decisions for sight-threatening retinal disease after training on 14,884 scans (PMID 30104768). Both used past scans, not new patients. What would prove more: prospective trials in screening programs.

AI co-scientist and C2S-Scale for drug ideas. Peer-reviewed and Company-reported, Preclinical. The co-scientist, built on Gemini, proposed drug repurposing candidates for acute myeloid leukemia that were confirmed in lab tests (PMID 42156544), and in human liver organoids at Stanford, one of its picks, the cancer drug vorinostat, cut TGF-beta induced chromatin changes by 91% (PMID 40946179). C2S-Scale, a 27 billion parameter model built with Yale, nominated silmitasertib to boost antigen presentation, and tests in human cell models supported it, but that work is still a preprint (PMID 41279114). What would prove more: animal studies, then a clinical trial of any of these ideas.

AlphaMissense and AlphaGenome for genetic variants. Peer-reviewed, Research tool. AlphaMissense classifies 89% of possible human missense variants as likely benign or likely pathogenic (PMID 37733863). AlphaGenome reads 1 million DNA letters at a time and matched or beat the best outside models in 25 of 26 variant effect tests (PMID 41606153). An independent Mount Sinai preprint found it improved personal gene expression prediction but kept key limitations (PMID 42039486). What would prove more: use in clinical genetic diagnosis with measured gains.

Who Google DeepMind Works With

Google DeepMind works mostly with hospitals, universities and public science bodies, and it routes drug making through its sister company Isomorphic Labs. Named partners on this page include the Royal Free London NHS trust (Streams), Moorfields Eye Hospital, EMBL-EBI (the AlphaFold database), Beth Israel Deaconess Medical Center (AMIE), Stanford (AI co-scientist liver work) and Yale (C2S-Scale). Thailand’s national diabetic retinopathy screening program worked with Google Health.

Isomorphic Labs is a separate Alphabet company launched in 2021. It co-developed AlphaFold 3 with Google DeepMind and raised $2.1 billion in May 2026, as Reuters reported. See our Isomorphic Labs drug discovery page. In June 2026 Google DeepMind lost AlphaFold co-creator John Jumper to Anthropic, as Bloomberg Government reported.

What to Watch Next

• A randomized trial of AMIE, or any Google conversational model, with patient outcome or diagnostic accuracy endpoints.
• Isomorphic Labs dosing its first human trial participant. Reuters reported in January 2026 that the company had pushed back its clinical trial timeline (Reuters via WHBL).
• Animal or human data on a drug candidate first proposed by the AI co-scientist or C2S-Scale.
• Peer-reviewed publication of the C2S-Scale preprint and of independent AlphaGenome evaluations.
• Any regulatory clearance for a MedGemma-based product built by an outside developer.

Related: our AI healthcare breakthroughs tracker, the Isomorphic Labs timeline, the Anthropic and Claude health timeline, the AI lab biotech deal tracker, our explainer on virtual cell AI and our HYPE Index methodology.

What This Means for You

If you are a patient, Google’s AI may already touch your care indirectly: AlphaFold structures sit behind a great deal of current lab research, and some screening programs use Google eye screening tools. But no drug designed with Google DeepMind’s tools has been approved, and chat tools like AMIE are still research systems that ran under physician supervision.

If you work in research or care, AlphaFold, AlphaMissense, AlphaGenome, TxGemma and MedGemma are tools you can use now. Treat their outputs as hypotheses to check, as the independent AlphaFold review advised, and ask whether a clinical model was tested on past data, simulated patients or real patients before you rely on it.

Compiled by Healthcare Discovery from company announcements, named news reports and PubMed. Every PMID on this page was checked against PubMed on October 28, 2026. Evidence levels follow the scale explained in our AI healthcare breakthroughs tracker. Healthcare Discovery uses AI research tools from several companies on this page, including OpenAI, Anthropic and Google models; no company listed reviewed or paid for this page.

Sources

1. Brodeur PG, et al. Conversational diagnostic artificial intelligence in ambulatory primary care: a prospective feasibility study. Lancet, 2026. PMID 42849491
2. Gottweis J, et al. Accelerating scientific discovery with Co-Scientist. Nature, 2026. PMID 42156544
3. Liévin V, et al. Towards conversational artificial intelligence for disease management. Nature, 2026. PMID 42310463
4. Avsec Z, et al. Advancing regulatory variant effect prediction with AlphaGenome. Nature, 2026. PMID 41606153
5. Shen L. AlphaGenome Enhances Personal Gene Expression Prediction but Retains Key Limitations. bioRxiv (preprint), 2026. PMID 42039486
6. Rizvi SA, et al. Scaling Large Language Models for Next-Generation Single-Cell Analysis. bioRxiv (preprint), 2026. PMID 41279114
7. Guan Y, et al. AI-Assisted Drug Re-Purposing for Human Liver Fibrosis. Advanced Science, 2025. PMID 40946179
8. Tu T, et al. Towards conversational diagnostic artificial intelligence. Nature, 2025. PMID 40205050
9. McDuff D, et al. Towards accurate differential diagnosis with large language models. Nature, 2025. PMID 40205049
10. Singhal K, et al. Toward expert-level medical question answering with large language models. Nature Medicine, 2025. PMID 39779926
11. Abramson J, et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature, 2024. PMID 38718835
12. Terwilliger TC, et al. AlphaFold predictions are valuable hypotheses and accelerate but do not replace experimental structure determination. Nature Methods, 2023. PMID 38036854
13. Cheng J, et al. Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science, 2023. PMID 37733863
14. Singhal K, et al. Large language models encode clinical knowledge. Nature, 2023. PMID 37438534
15. Ruamviboonsuk P, et al. Real-time diabetic retinopathy screening by deep learning in a multisite national screening programme: a prospective interventional cohort study. Lancet Digital Health, 2022. PMID 35272972
16. Varadi M, et al. AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models. Nucleic Acids Research, 2022. PMID 34791371
17. Jumper J, et al. Highly accurate protein structure prediction with AlphaFold. Nature, 2021. PMID 34265844
18. Tunyasuvunakool K, et al. Highly accurate protein structure prediction for the human proteome. Nature, 2021. PMID 34293799
19. Senior AW, et al. Improved protein structure prediction using potentials from deep learning. Nature, 2020. PMID 31942072
20. McKinney SM, et al. International evaluation of an AI system for breast cancer screening. Nature, 2020. PMID 31894144
21. De Fauw J, et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nature Medicine, 2018. PMID 30104768
22. Bloomberg Government. Nobel Winner John Jumper to Leave Google DeepMind for Anthropic, June 2026. Bloomberg Government
23. Reuters via WHBL. Google-backed AI drug discovery startup Isomorphic Labs delays clinical trial timeline, January 2026. WHBL
24. Reuters via WHTC. Google-backed Isomorphic raises $2.1 billion to scale AI-driven drug discovery, May 2026. WHTC
25. Google. How a Gemma model helped discover a new potential cancer therapy pathway, October 2025. Google
26. InfoQ. DeepMind AlphaGenome, July 2025. InfoQ
27. HTN. Google launches new open model for health app developers, May 2025. HTN
28. Google for Developers. Introducing TxGemma: Open models to improve therapeutics development, March 2025. Google for Developers
29. PharmTech. Google launches AI co-scientist system, February 2025. PharmTech
30. Nobel Prize. Press release: The Nobel Prize in Chemistry 2024, October 2024. Nobel Prize
31. EMBL. AlphaFold 200 million structures, July 2022. EMBL
32. UPI. Alphabet launches Isomorphic Labs to use artificial intelligence to find drugs, November 2021. UPI
33. Bio-IT World. DeepMind Releases Open Protein Structure Database Including Complete Human Proteome, July 2021. Bio-IT World
34. DeepMind. AlphaFold: a solution to a 50-year-old grand challenge in biology, November 2020. DeepMind
35. Synced. DeepMind AlphaFold Delivers Unprecedented Progress on Protein Folding, December 2018. Synced
36. DeepMind. Scaling Streams with Google, November 2018. DeepMind
37. Digital Health. Royal Free and Google DeepMind trial did not comply with DPA, July 2017. Digital Health
38. DeepMind. Working with the NHS to build lifesaving technology, November 2016. DeepMind

Free Daily Briefing

The Latest Longevity Science.
Delivered Every Morning.

Join researchers, physicians, and health professionals getting daily breakthroughs in AI-driven medicine, epigenetics, and longevity research.

Support the research that powers this editorial

Subscribe Free

No spam. Unsubscribe anytime. We respect your inbox.

Keep Reading on Healthcare Discovery

• The 12-Hour Calm: What the 2025 Cold Water Immersion Evidence Really Says About Stress, Sleep, and Recovery
• The 95 Percent Warning: What Healthcare AI Must Learn From Prior Authorization Appeals
• The VO2 Max Imperative: What 50 Years of Research Reveal About the Strongest Mortality Predictor in Modern Medicine