Research AI Landscape

alphafold

3 min read

AlphaFold

Status: adopt | Maturity: production-ready | Last reviewed: 2026-03-20

What it does

AlphaFold is an AI system from Google DeepMind that predicts the three-dimensional structure of proteins from amino acid sequences. AlphaFold2 (released 2021) achieved near-experimental accuracy across the majority of known protein families and was recognised as a breakthrough contributing to the 2024 Nobel Prize in Chemistry. The AlphaFold Protein Structure Database, hosted by EMBL-EBI, provides pre-computed structures for over 200 million proteins.

AlphaFold3 (released 2024) extends predictions to protein–DNA, protein–RNA, and protein–small molecule complexes, enabling drug discovery and biochemical interaction modelling.

Key capabilities

  • Accurate protein structure prediction from sequence alone
  • Pre-computed database of 200 million+ protein structures (AlphaFold DB, EMBL-EBI)
  • AlphaFold3: multi-molecule complex prediction (proteins, DNA, RNA, ligands)
  • Confidence scoring per residue (pLDDT) to flag uncertain regions
  • Open-source code for local deployment with custom datasets
  • AlphaFold Server for web-based predictions without local compute

Fit for Sheffield researchers

Essential resource for structural biology, biochemistry, pharmacology, and related health/science research. The EMBL-EBI database is free and directly accessible to any researcher. Sheffield faculty in the School of Biosciences, Chemistry, and Health Sciences are likely already using AlphaFold — this note codifies support and governance guidance.

Local deployment (for novel predictions not in the database) requires significant GPU compute. Research IT / CiCS should assess whether Sheffield’s HPC cluster (Stanage) is a viable route for AlphaFold3 runs.

Licensing and access

  • AlphaFold DB: free to access at https://alphafold.ebi.ac.uk (CC BY 4.0)
  • AlphaFold Server: free for non-commercial research
  • AlphaFold2 code: Apache 2.0 open source
  • AlphaFold3 model weights: available to academic researchers under a more restrictive licence (no commercial use, weights must be applied for separately from Google DeepMind)

Trust and governance notes

The EMBL-EBI database hosts pre-computed structures — no data upload required for most use cases. AlphaFold Server uploads sequences to Google infrastructure; researchers should not upload unpublished or commercially sensitive sequences without reviewing Google DeepMind’s terms. Data residency for the EMBL-EBI portal is EU (UK/Germany).

No personal data is typically involved in protein structure work, reducing GDPR exposure for standard use cases.

Observed limitations

  • Predictions for novel protein families with few known homologues remain less reliable
  • AlphaFold3 nucleic acid and small molecule predictions are less validated than protein predictions
  • Local AlphaFold3 deployment requires significant GPU resources and technical expertise
  • Model weights for AlphaFold3 are restricted (academic licence only, must apply)
  • Does not predict protein dynamics or conformational changes (static structures only)

chemcrow · labos · deepsomatic

Sources and evidence

Graph View

Backlinks