EXPLAINER

AlphaFold (Protein Structure Prediction)

modelfreshness: volatilesources not yet verified

AlphaFold is a family of deep-learning models from Google DeepMind that predict protein 3-D structure from amino-acid sequence with accuracy competitive with experiments for many proteins.

Short answer: AlphaFold 2 (2020–2021) predicted single protein structures with high accuracy for many targets; AlphaFold 3 (2024) extends to complexes with other molecules. Predictions come with confidence scores and are models, not experimental structures — low-confidence regions and dynamics need caution.

What is it?

Proteins are chains of amino acids that fold into complex 3-D shapes which determine their function. AlphaFold predicts that shape from the sequence, which used to take months or years of lab work per protein. The predictions are very useful but still need care, especially where the model is not confident.

Also known as: AlphaFold, AlphaFold2, AlphaFold 3, AlphaFold database, protein structure prediction, protein folding AI, pLDDT, ESMFold.

Technical definition

AlphaFold 2 (Jumper et al., Nature, 2021) used an attention-based architecture (Evoformer plus a structure module) with multiple-sequence-alignment and template inputs, producing per-residue confidence (pLDDT) and predicted aligned error. The AlphaFold Protein Structure Database made hundreds of millions of predictions openly available (verify current scale and hosting, EMBL-EBI). AlphaFold 3 (Abramson et al., Nature, 2024) uses a diffusion-based approach to predict joint structures of proteins, nucleic acids, ligands and ions; its code is published on GitHub, but the model weights are provided separately under their own terms of use, and the AlphaFold Server offers non-commercial use (access terms as of 2026-10-09). The 2024 Nobel Prize in Chemistry recognised Demis Hassabis and John Jumper (for AlphaFold) and David Baker (computational protein design). Related models include ESMFold and RoseTTAFold.

Entity type: model · Freshness class: volatile.

When should I use it?

Useful hypothesis generator in structural biology and drug-discovery pipelines.

Implementation patterns

  • Look up existing predictions before running your own.
  • Filter by pLDDT/PAE before using a region.
  • Validate key hypotheses experimentally.
  • Use multiple predictors for difficult targets.

Technologies and frameworks

Commonly associated: AlphaFold DB (EMBL-EBI), ColabFold, ESMFold, RoseTTAFold, OpenFold. Names are examples of what exists, not recommendations, and projects change quickly.

Practical guide

  1. Search the database for your protein.
  2. Read pLDDT and PAE.
  3. Cross-check with literature.
  4. Plan experimental validation.
  5. Check terms for the model version you use.

Common failure modes

  • Treating low-confidence regions as real structure.
  • Assuming a single static structure captures dynamics.
  • Using predictions where mutations/ligands change the fold.
  • Ignoring licence terms of specific model versions.

Troubleshooting

  • Disordered or low-confidence regions — Interpret as uncertain; consider experimental data or ensembles.
  • Complex prediction implausible — Check PAE between chains; compare other tools or experiments.

Contrasted with

  • AI in Drug Discovery — AlphaFold predicts structure; drug discovery also needs binding, safety and clinical validation.

Questions this page answers

  • What is AlphaFold?
  • How does AlphaFold work?
  • What does pLDDT mean?
  • Is AlphaFold accurate?
  • What is the AlphaFold database?

Typical goals: understand how protein structure prediction works; interpret AlphaFold confidence scores.

Sources and verification

Status: not independently verified. This entry was authored on 2026-10-06 from general technical knowledge, without live checking of the sources below. Treat every source as a lead to confirm (identifier, title, authorship, current version) before publication or citation.

Freshness class: volatile — Products, versions, protocols, benchmarks, leaderboards, regulations and dates. Verify against the primary source before relying on specifics; re-review every 3 months.

Time-sensitive claims to verify before relying on them:

  • Database size, access terms and licences for model versions.
  • Nobel Prize 2024 details.
  • Current successor models and benchmarks.

Canonical sources (unverified leads):

  • Jumper et al., Highly accurate protein structure prediction with AlphaFold, Nature 596 (2021) — doi:10.1038/s41586-021-03819-2
  • Abramson et al., Accurate structure prediction of biomolecular interactions with AlphaFold 3, Nature (2024) — doi:10.1038/s41586-024-07487-w
  • AlphaFold Protein Structure Database — alphafold.ebi.ac.uk (verify)