Cancer3.AI › AI in Oncology › AI Models › AlphaFold 2
Foundation model Open weights Not applicable protein-structure

AlphaFold 2 2

Predicts the 3D structure of a protein from its amino-acid sequence at near-experimental accuracy; the AlphaFold Protein Structure Database (with EMBL-EBI) provides predicted structures for over 200 million proteins, including cancer-relevant targets and mutants.

GitHub Paper DOI Demo { } JSON

At a glance

DeveloperGoogle DeepMind
Version2
Released2021-07-15
LicenceApache-2.0 (code) / CC-BY-4.0 (parameters)
AvailabilityOpen weights
KindFoundation model
Regulatory statusNot applicable
Parameters~93M

What it does

Sequence in, atomic coordinates and per-residue confidence (pLDDT) out. In oncology it underpins structural interpretation of driver mutations, target druggability assessment and structure-based drug design.

Notable uses in oncology

CASP14 winner (2020); more than 30% of publications citing AlphaFold concern disease research, including analyses of mutated cancer proteins; feeds Isomorphic Labs' drug-design work.

Tasks, data types and cancers

TaskProtein structure
Data typeProtein sequence
Clinical settingBasic researchDrug discovery
CancerPan-cancer
Inputamino-acid sequence (+ multiple-sequence alignment, templates)
Output3D coordinates, pLDDT, PAE

Architecture

FamilyEvoformer + structure module (attention over MSA and pair representations)
Parameters~93M
Pre-trainingsupervised on PDB structures + self-distillation on predicted structures

Training data

Experimentally determined structures from the Protein Data Bank (cut-off 2018) plus self-distillation on predicted structures of UniRef sequences.

InstitutionsDeepMind; PDB, UniProt as data sources

Linked datasets

Evaluation

Benchmark / datasetMetricValueExternal validationSource
CASP14 GDT_TS median 92.4 across targets yes Source

How to run

# fastest path: ColabFold (AlphaFold2 + MMseqs2)
pip install colabfold[alphafold]
colabfold_batch input.fasta out_dir/
# or query precomputed structures: https://alphafold.ebi.ac.uk/entry/<UniProt accession>
LibraryJAX (repo) / ColabFold
HardwareGPU; MSA search is the slow part — ColabFold with MMseqs2 is the practical route

Regulatory status and intended use

Regulatory statusNot applicable
Intended useResearch tool.

Source →

Regulatory status is quoted from the source linked above and can change. Research-use-only models must not be used for clinical decisions.

Limitations and bias

  • Static single conformation; poor for intrinsically disordered regions and many point-mutation effects.
  • No ligands or complexes with nucleic acids in AlphaFold 2 (see AlphaFold 3).

Sources

  1. Jumper J et al. Highly accurate protein structure prediction with AlphaFold. Nature 2021
  2. AlphaFold Protein Structure Database
  3. Google DeepMind — AlphaFold

This page is educational — it is not medical advice and does not replace consultation with an oncologist. Diagnostic and treatment decisions are made solely by specialist physicians.