Database

AlphaSync currently covers 2,617,720 proteins across 925 species, prioritizing model organisms and global health proteomes from the AlphaFold Protein Structure Database and the 200 representative eukaryotic species included in the Ensembl Comparative Genomics pipeline. The current source database versions used are AlphaFold DB v4 (Oct 2022), UniProt 2026_01 and Ensembl 108 (Oct 2022). AlphaSync contains all up-to-date reviewed and canonical reference proteome proteins for 43 “completed” species. Beyond these species, while proteomes may not be complete, all proteins contained in AlphaSync are up-to-date with UniProt. Please see the Browse section for a full list of these. It also contains the complete UniProt set of human proteins, but is limited in Ensembl Comparative Genomics species to only include proteins that are high-confidence one–to–one orthologs or “best” one–to–many orthologs of human proteins to keep the size of the dataset manageable. The current dataset size is 447 GB across 9 SQL tables. A convenient REST API is available for automated data access.

Methods

AlphaFold 2 structure predictions are obtained from the AlphaFold Protein Structure Database. Residue solvent accessibility is calculated using DSSP 4.2.2.1, with relative solvent accessibilities calculated using the reference values from Tien et al. in PLOS One (2013). Residues that are ≤25% exposed according to their relative solvent-accessible surface area (RSA) are considered buried, as defined by Levy et al. an JMB (2010), while residues with RSA >25% are considered surface residues. Residues with an average RSA across a ±10 residue window (RSA₁₀) of ≥55% are considered to be part of an intrinsically disordered region, as described in Akdel et al. in NSMB (2022), while residues with RSA₁₀ <55% are considered structured. Fragment structures for human proteins longer than 2,700 amino acids are combined by averaging e.g. ASA, RSA and pLDDT scores, while ignoring values within 200 aa of artificial protein termini introduced by fragmentation. Proteins are fragmented into 1400 aa segments beginning at the N-terminus and stepping by 200 aa, as described by Tunyasuvunakool et al. in Nature (2021) for the human proteome only. For categorical values such as secondary structure, the most common structure type for a given residue is reported, using the pLDDT score as a tiebreaker. Contacts are detected using the Lahuta Python package, also developed at St. Jude Children’s Research Hospital by Dr. Besian Sejdiu. When selecting proteins for inclusion in species other than human, which is fully represented, all high-confidence one-to-one orthologs to human proteins are included, as well as a single “best” one-to-many ortholog for each species that is chosen based on whole-genome alignment, synteny and sequence identity. This ensures that each of the 200 representative eukaryotic species included in the Ensembl Comparative Genomics pipeline is represented by at most a single protein in a given homology cluster, and keeps the size of the AlphaSync database manageable. Web combonents: Bootstrap, Font Awesome for icons, bootstrap-sortable for sortable tables, and ExcellentExport.js for table export to CSV files.

How to cite / reference

Please cite our NSMB paper: Lang, B. et al. AlphaSync is an enhanced AlphaFold structure database synchronized with UniProt. Nat Struct Mol Biol (2025). https://doi.org/10.1038/s41594-025-01719-x.

Credits

Graphic design (e.g. logo) by Nadia Arancibia Villaleiva.

Contact

Please contact Benjamin Lang and M. Madan Babu via email. Any ideas, questions and feedback are very welcome!