The AlphaFold AI system developed by DeepMind predicts a protein’s 3D structure from its amino acid sequence. You can now view variant locations on AlphaFold predicted structures in the Ensembl VEP web tool.

Knowing where a variant falls in a protein structure is important for interpreting its potential impact. To help with this, Ensembl VEP has displayed your variants on experimentally derived PDBe structures since 2019. As it is complex to derive protein structures experimentally, they are only available for a fraction of genes. The AlphaFold database which now holds 200 million largely high quality protein structure predictions, including primary isoforms for the greater majority of human protein coding genes has changed what is possible.

To view variants on protein structures, select the ‘Protein Matches’ option on the VEP interface. Variants are mapped to structures as part of the Ensembl VEP analysis, so novel as well as previously known variants can be viewed. In the results table, click on the ’AlphaFold Model’ button in the ‘Protein Matches’ column to launch the interactive protein structure visualisation. Exons and protein domain annotation can optionally be displayed to provide more context. We also show the confidence score for each residue as a guide to reliability.

Ensembl VEP results for mouse variant NC_000067.7:4898856:A:C, showing the variant in purple on the AlphaFold predicted protein structure.

This visualisation is currently available in Ensembl VEP analysis of human and mouse missense variants. It will be enabled for further species and variant types in coming Ensembl releases.