AlphaFold Database release notes

AlphaFold version 6 release notes

 

 

 

 

 

 

 

 

 

 

 

 

We’ve released AlphaFold DB v6, a major update that brings the database in sync with UniProt release 2025_03. This new version significantly expands the coverage and ensures the database is aligned with the latest reference proteomes.

With this release, we've added over 26 million new predicted protein structures, bringing the total to more than 241 million. 

Key statistics

  • AlphaFold DB v4: 214,683,829 predicted protein structures
  • AlphaFold DB v6: 241,070,489 predicted protein structures  (including 40,054 isoforms)

Differences in the number of proteins since v4:

  • Added: 65,711,653
  • Removed: 39,324,993
  • Changed: 276,539
  • Unchanged: 175,082,297

Interpretation:

  • “Added” and “Removed” follow upstream UniProt: new proteins predicted, and records archived upstream are retired. “Added” includes isoforms too.
  • “Changed” refers to amino-acid sequence changes (not metadata).
  • “Unchanged” entries received metadata updates to reflect current UniProt information and are relabelled v6. The underlying model coordinates have not been updated, but carried over from v4.

For a detailed list of UniProt accessions for each category, check our FTP page at: https://ftp.ebi.ac.uk/pub/databases/alphafold/. 

  • diffs.ndjson.gz (newline-delimited JSON). See README.txt there for field definitions.
  • Accession_ids.csv: Current accessions (v6) in the database

Accessing AlphaFold data

The AlphaFold Database website exclusively serves the most current, up-to-date models. If you need to access older versions, such as v4, they are available in our FTP area as complete proteomes at: https://ftp.ebi.ac.uk/pub/databases/alphafold/. 

Please note that the old URL format, such as https://alphafold.com/entry/P40763, no longer takes you directly to a specific model page. The new, preferred URL format is https://alphafold.com/entry/AF-P40763-F1.

For programmatic access to the most current files (PDB, CIF, or bCIF), we highly recommend using the AlphaFold API. This is the most reliable way to get download URLs for the latest available file versions. You can find the API documentation here: https://alphafold.com/api-docs.

API changes

We recently announced upcoming changes to the "prediction" API endpoint, primarily driven by our expanding dataset. To help users transition, we've created a carefully planned API deprecation timeline.

We are currently in a Dual Support Period that will last for 9 months, during which both the old and new API fields will be supported. Please update your workflows as soon as possible. The sunset date for the old API fields is 25-06-2026.

For a full list of deprecated and new fields, please see our dedicated API news item: https://www.ebi.ac.uk/pdbe/news/breaking-changes-afdb-predictions-api. 

For a complete schema reference, visit https://alphafold.ebi.ac.uk/api-doc. 

Multiple Sequence Alignments

  • MSAs used to predict each model are available on individual entry pages.
  • Bulk access is not currently planned.

Feedback

We value your input. If you have any issues or suggestions, please contact us via our Helpdesk at afdbhelp@ebi.ac.uk.