Subscribe in a reader

Cambridge MedChem Consulting

AlphaFold Protein Structure Database in 2024

A recent publication describes the continued evolution of the AlphaFold Protein Structure Database created by EMBL-EBI and DeepMind. From an initial 300K structures it now contains 214 million predicted protein structures.

You can read the paper here DOI.

The AlphaFold Database Protein Structure Database (AlphaFold DB, has significantly impacted structural biology by amassing over 214 million predicted protein structures, expanding from the initial 300k structures released in 2021. Enabled by the groundbreaking AlphaFold2 artificial intelligence (AI) system, the predictions archived in AlphaFold DB have been integrated into primary data resources such as PDB, UniProt, Ensembl, InterPro and MobiDB. Our manuscript details subsequent enhancements in data archiving, covering successive releases encompassing model organisms, global health proteomes, Swiss-Prot integration, and a host of curated protein datasets. We detail the data access mechanisms of AlphaFold DB, from direct file access via FTP to advanced queries using Google Cloud Public Datasets and the programmatic access endpoints of the database. We also discuss the improvements and services added since its initial release, including enhancements to the Predicted Aligned Error viewer, customisation options for the 3D viewer, and improvements in the search engine of AlphaFold DB.

AlphaFold predicts structure of almost every catalogued protein known to science

A little over a year ago I highlighted the AlphaFold Protein Structure Database in which AlphaFold DB provided open access to protein structure predictions for the human proteome and 20 other key organisms to accelerate scientific research. Well things have moved on.

DeepMind and EMBL’s European Bioinformatics Institute (EMBL-EBI) have made AI-powered predictions of the three-dimensional structures of nearly all catalogued proteins known to science freely and openly available to the scientific community, via the AlphaFold Protein Structure Database.

The database is being expanded by approximately 200 times, from nearly 1 million protein structures to over 200 million, covering almost every organism on Earth that has had its genome sequenced. The expansion of the database includes predicted structures for a wide range of species, including plants, bacteria, animals, and other organisms.

The full dataset of all predictions is available at no cost and under a CC-BY-4.0 licence from Google Cloud Public Datasets. We've grouped this by single-species for ease of downloading subsets or all of the data. We suggest that you only download the full dataset if you need to process all the data with local computing resources (the size of the dataset is 23 TiB, ~1M tar files).

Downloads can be found here

It is worth noting that AlphaFold2 is not the only protein structure prediction tool available, there is also RoseTTAFold, OpenFold, and FastFold.

CASP15 details

The details of the latest Critical Assessment of Structure Prediction (CASP) experiment to determine and advance the state of the art in modeling biomolecular structures have been published

Modeling categories

The core of CASP remains the same: blind testing of methods with independent assessment against experiment to establish the state-of-art in modeling proteins and protein complexes. CASP15 will include following categories.

  • Single Protein and Domain Modeling As in previous CASPs, the accuracy of single proteins and where appropriate single protein domains will be assessed, using the established metrics. Two changes will be the elimination of the distinction between template-based and template-free modeling, and an emphasis on the fine-grained accuracy of models, such as local main chain motifs and side chains. Because of the high accuracy of the new modeling methods, we expect assessment against high resolution experimental structures will be most informative.
  • Assembly As in recent CASPs, the ability of current methods to correctly model domain-domain, subunit-subunit, and protein-protein interactions will be assessed. We will again work in close collaboration with our CAPRI partners. Because of the promising deep learning results reported so far, substantial progress is expected.
  • Accuracy Estimation Members of the community will be invited to submit accuracy estimates for multimeric complexes and inter-subunit interfaces. There will no longer be a category for estimating the accuracy of single protein models, since it has become clear these cannot compete with modeling method specific estimates. Instead, there will be increased emphasis on assessment of self-reported accuracy estimates at the atomic level. Note the units will now be pLDDT, not Angstroms.
  • RNA structures and complexes There will be a pilot experiment to assess the accuracy of modeling for RNA models and protein-RNA complexes. The assessment will be done in collaboration with the RNA-Puzzles and Marta Szachniuk's group in Poznan.
  • Protein-ligand complexes Subject to the availability of adequate resources, there will also be a pilot experiment in this area. Deep-learning is already having an impact here, and there is high interest because of the relevance to drug design.
  • Data Assisted As in recent CASPs, there will be assessment of the extent to which the accuracy of models can be increased by the provision of sparse data, particularly that provided by SAXS and mass spectroscopy/chemical crosslinking. Only targets where these low-resolution data are likely to be useful will be considered, that is, large single proteins and complexes. As previously, we will work with collaborators to obtain the necessary experimental data. Targets will initially be released without the experimental data, followed by a second round of prediction including those data.
  • Protein conformational ensembles Following the success of deep-learning methods for single structures, it is increasingly important to assess methods for predicting structure ensembles. This is a huge area, ranging from the many conformations of disordered regions to the small number of conformations that may be involved in allosteric transitions and enzyme excited states to local protein dynamics. While it is clear that deep learning and other methods have the potential to generate ensembles in some circumstances, the difficulty is in finding cases where there are sufficiently accurate and extensive experimental data to allow rigorous assessment. One promising avenue is modeling sets of conformations in regions of cryo-EM structures where there is evidence of local conformational heterogeneity. If suitable cases arise, we will present these as a special type of sub-target. First requesting conformational ensembles that will be evaluated against the election density map and then in a possible second stage providing the map for data assisted ensemble prediction. A second possibility is for cases where detailed NMR data have already established the structure of two or more conformations. We have a good lead for a few targets of this type. In addition to this, we are considering a non-blind experiment (a departure from normal CASP practice), where we will first ask those interested to reproduce the known conformations. We will also ask participants to identify any additional conformations that appear to be present. It may then be possible to test these against existing or new experimental data.

Details of the targets will be made available over the next week

AlphaFold Protein Structure Database

The AlphaFold Protein Structure Database Developed by DeepMind and EMBL-EBI is now available online.

AlphaFold DB provides open access to protein structure predictions for the human proteome and 20 other key organisms to accelerate scientific research.

AlphaFold DB currently provides predicted structures for the organisms listed below and includes human, laboratory species, and key pathogens. All the predictions for all the species can be downloaded from the EBI FTP site

Species Common Name Reference Proteome Predicted Structures Download
Arabidopsis thaliana Arabidopsis UP000006548 27,434 Download (3642 MB)
Caenorhabditis elegans Nematode worm UP000001940 19,694 Download (2601 MB)
Candida albicans C. albicans UP000000559 5,974 Download (965 MB)
Danio rerio Zebrafish UP000000437 24,664 Download (4141 MB)
Dictyostelium discoideum Dictyostelium UP000002195 12,622 Download (2150 MB)
Drosophila melanogaster Fruit fly UP000000803 13,458 Download (2174 MB)
Escherichia coli E. coli UP000000625 4,363 Download (448 MB)
Glycine max Soybean UP000008827 55,799 Download (7142 MB)
Homo sapiens Human UP000005640 23,391 Download (4784 MB)
Leishmania infantum L. infantum UP000008153 7,924 Download (1481 MB)
Methanocaldococcus jannaschii M. jannaschii UP000000805 1,773 Download (171 MB)
Mus musculus Mouse UP000000589 21,615 Download (3547 MB)
Mycobacterium tuberculosis M. tuberculosis UP000001584 3,988 Download (421 MB)
Oryza sativa Asian rice UP000059680 43,649 Download (4416 MB)
Plasmodium falciparum P. falciparum UP000001450 5,187 Download (1132 MB)
Rattus norvegicus Rat UP000002494 21,272 Download (3404 MB)
Saccharomyces cerevisiae Budding yeast UP000002311 6,040 Download (960 MB)
Schizosaccharomyces pombe Fission yeast UP000002485 5,128 Download (776 MB)
Staphylococcus aureus S. aureus UP000008816 2,888 Download (268 MB)
Trypanosoma cruzi T. cruzi UP000002296 19,036 Download (2905 MB)
Zea mays Maize UP000007305 39,299 Download (5014 MB)

The search bar at the top of the query page accepts queries based on protein name, gene name, UniProt identifier, or organism name. At present you can't search using a sequence and look for similar proteins. You would first need to do a BLAST search and use the results from that as queries.

Here I searched for Plasmodium falciparum carbonic anhydrase (Q8IHW5) a potential Malaria target. As you can see there is no crystal structure in the PDB. Whilst the active site is predicted with high confidence there are clearly regions for which there is very low confidence.


You can then download the structure in PDB or mmCIF format.

I made a homology model (in purple below) of this protein a while back and it has little sequence similarity with any proteins in the PDB. Despite not including a Zinc the Alphafold Predicted Structure includes histidines in positions to potentially coordinate to the Zinc. If it is possible to include the Zinc in the structure prediction I'd be interested in finding out.


Overall I'd say this is a very useful starting point.