Tech & AI, Explained

alphafold

4 min read

🇬🇧 UK · DeepMind AlphaFold (and AlphaProof / AlphaGeometry / AlphaEarth)

Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: Google DeepMind’s scientific-discovery AI family. AlphaFold solved protein-structure prediction (Nobel Prize 2024). AlphaProof + AlphaGeometry tackle mathematical proofs. AlphaEarth predicts Earth-system dynamics. The most impactful “AI for science” work to date.

Front-matter facts

FieldValue
VendorGoogle DeepMind (London, UK — DeepMind is a UK-founded lab, acquired by Google 2014)
Country / origin🇬🇧 UK + 🇺🇸 USA (DeepMind founded London 2010; now part of Google)
Recommended for Australian users?✅ Yes — AlphaFold database is free + globally accessible; research-focused
Privacy summaryResearch tools; AlphaFold predictions are public; no personal-data implications for typical use
Free tierAlphaFold database + tools are free for non-commercial use; Nobel-Prize-winning science freely shared
Paid tiersSome commercial-use restrictions; Isomorphic Labs spinoff handles commercial drug discovery
First releasedAlphaFold 2014; AlphaFold 2 (breakthrough) July 2020; AlphaFold 3 May 2024
Last reviewed2026-06-26
Official sitehttps://deepmind.google/technologies/alphafold/

What it is

The Alpha-series is Google DeepMind’s family of AI systems for scientific discovery and mathematical reasoning. They’re not consumer products — they’re research breakthroughs that have transformed specific scientific fields.

AlphaFold (most-famous)

  • Predicts protein 3D structure from amino-acid sequence
  • AlphaFold 2 (2020) solved a 50-year-old grand-challenge problem in biology
  • AlphaFold 3 (May 2024) extended to protein-DNA / RNA / small-molecule interactions — broader drug-discovery applications
  • AlphaFold Database has ~200 million predicted protein structures, free for non-commercial use
  • Demis Hassabis + John Jumper won the 2024 Nobel Prize in Chemistry for AlphaFold

AlphaProof

  • Solves Mathematical Olympiad problems
  • Combines Gemini language model with formal proof verification (Lean theorem prover)
  • 2024: solved 4 of 6 IMO 2024 problems at silver-medal level

AlphaGeometry

  • Solves geometry Olympiad problems
  • Often paired with AlphaProof for full IMO-level reasoning

AlphaEarth

  • Earth-system AI — climate, weather, ecology predictions
  • Newer; less mature than AlphaFold

AlphaTensor, AlphaCode, AlphaDev, AlphaMissense, AlphaQubit, etc.

  • DeepMind’s broader “Alpha” series — algorithms, code, biology, physics, quantum

Commercial spin-off

  • Isomorphic Labs — Alphabet’s drug-discovery company, uses AlphaFold + related AI for therapeutics development

What you’d use it for

AlphaFold (most accessible)

  • Biology / biochemistry researchers — predict protein structures for any sequence
  • Drug discovery — understand how proteins interact (basic research)
  • Education — learn about protein structure with visualisations
  • Bioinformatics — incorporate AlphaFold predictions into pipelines

AlphaProof / AlphaGeometry

  • Mathematical research — formal verification of proofs
  • Education — see how AI tackles competition-level mathematics

General significance

  • Even if you don’t use these directly, AlphaFold’s free database has accelerated research worldwide — most modern biology / drug discovery references it

How to access AlphaFold from Australia

AlphaFold Database (free)

  • Go to alphafold.ebi.ac.uk
  • Search for any protein (by sequence, name, or UniProt ID)
  • Download structure, view in browser
  • ~200M predicted structures available

AlphaFold Server (for new sequences)

  • alphafoldserver.com — submit a sequence, get prediction
  • Free for non-commercial use
  • AlphaFold 3 capabilities

Code / models

  • Open-source releases on GitHub (deepmind/alphafold)
  • For computational researchers wanting to run locally

What it costs

AlphaFold Database + Server

  • Free for non-commercial use
  • Commercial use requires permission / Isomorphic Labs commercial license

AlphaProof / AlphaGeometry / others

  • Mostly research-only; not consumer products
  • Code occasionally open-sourced

Drug discovery commercial

  • Via Isomorphic Labs partnerships (enterprise commercial terms)

How it compares

There’s no real direct competitor to AlphaFold — it’s a distinct research-AI achievement:

AspectAlphaFoldOther protein-structure toolsGenerative AI for science
Protein structure prediction accuracyBest (Nobel-level)Older / less accurateLimited / specialised
Free public accessYes (200M structures)LimitedVaries
Commercial potentialHigh (via Isomorphic Labs)Established pharmaEmerging

For science-AI as a category, DeepMind’s Alpha-series is the gold standard.

Privacy / data handling

  • Protein sequences submitted to AlphaFold Server: standard DeepMind / Google research terms
  • For sensitive intellectual property (novel drug candidates), commercial-license / Isomorphic-Labs route
  • AlphaFold predictions for public-database proteins are PUBLIC

Recent changes

  • 2024: Nobel Prize in Chemistry for AlphaFold (Hassabis + Jumper)
  • May 2024: AlphaFold 3 released
  • July 2024: AlphaProof + AlphaGeometry IMO silver-medal
  • 2020: AlphaFold 2 breakthrough (CASP14)
  • 2014: DeepMind acquired by Google

Gotchas

  • AlphaFold is for proteins, not general AI — don’t confuse with Gemini consumer chat
  • AlphaFold predictions are predictions — for high-stakes science, experimental validation still needed
  • Commercial use requires permission — Isomorphic Labs is Alphabet’s drug-discovery commercial arm
  • AlphaProof / Geometry / Tensor are research milestones, not products you’d casually use
  • Demis Hassabis (DeepMind CEO) is widely respected; DeepMind’s culture is research-first vs Google’s product focus
  • For Aussie researchers in biology, AlphaFold is essential infrastructure — likely already part of your tooling

See also

Sources

Graph View

Backlinks