AlphaMissense data integrated into Ensembl, UniProt, DECIPHER and AlphaFold DB
The integration enables researchers to easily access AI-generated scores estimating how likely genetic variants are to be pathogenic, and explore which protein regions have higher pathogenicity scores
Credit: Karen Arnott / EMBL-EBI
Summary
Knowing which genetic variants are pathogenic could help improve patient diagnosis and develop better treatments.
The AlphaMissense AI model, developed by Google DeepMind, classifies missense variants, predicting whether they are more likely to be pathogenic or benign.
AlphaMissense data is now integrated into several EMBL-EBI resources including Ensembl, UniProt, ProtVar and AlphaFold database.
Understanding which genetic variants are associated with diseases could help improve patient diagnoses and the development of better treatments. However, due to the wide range of possible genetic mutations, linking specific variants to disease remains challenging.
The AlphaMissense AI model, developed by Google DeepMind in 2023, classifies missense variants, predicting whether they are more likely to be pathogenic or benign. AlphaMissense is based on the AlphaFold 2 AI system, which predicts structures for nearly all proteins known to science from their amino acid sequences.
AlphaMissense analyses related protein sequences and the structural context of variants to estimate how likely a variant is to be pathogenic. AlphaMissense does not predict impact on biophysical properties or specific diseases. Regions of high pathogenicity scores within proteins are also indicative of evolutionary constraint or functional importance.
What is a missense variant?
A missense variant is a genetic alteration in which base pair substitution alters the genetic code to produce an amino acid that is different from the most commonly found amino acid at that position. The average person carries more than 9,000 missense variants. Most are benign and have little to no effect, but others are pathogenic and can severely disrupt protein function. Source: NIH – National Cancer Institute and Google DeepMind
Integration for easier access
To enable researchers to explore which regions of the protein have higher pathogenicity scores and easily obtain scores for variants of interest, EMBL’s European Bioinformatics Institute (EMBL-EBI) has integrated the AlphaMissense data into several of the open data resources it manages.
“We’re really excited to integrate AlphaMissense predictions into the EMBL-EBI tools that thousands of researchers use daily,” said Žiga Avsec, Research Scientist at Google DeepMind. “Overlaying AlphaMissense predictions onto AlphaFold structures can provide additional insights into protein function, such as highlighting potential binding interfaces. We hope that this integration will improve the accessibility of AlphaMissense and provide researchers with an extra piece of information about missense variants and their context.”
AlphaMissense has been integrated into the following freely accessible resources and tools:
Ensembl, the genome browser that supports research in comparative genomics, evolution, sequence variation and transcriptional regulation. AlphaMissense scores are integrated into the Ensembl Variant Effect Predictor tool, enabling easy annotation of variants via its user-friendly web interface, REST API or command line interface. Average AlphaMissense pathogenicity scores for each amino acid can also be visualised on the AlphaFold predicted 3D protein structure, available from the associated Ensembl transcript page. The interactive view allows switching between variants, domains, exons and AlphaMissense results to support interpretation of different regions of the protein and their sensitivity to change.
DECIPHER is a platform for phenotype-associated variant data sharing. It includes a suite of tools and visualisations which provide detailed information on genomic variants. AlphaMissense scores and pathogenicity categorisations are now displayed on DECIPHER variant annotation pages alongside other key information to further accelerate variant interpretation in rare disease investigations.
UniProt, the world-leading high-quality, comprehensive resource of protein sequence and functional information. Scientific literature is a rich source of known information on missense variants and their functional effects on disease phenotypes. All this is collected through expert manual curation in UniProt. Because AlphaMissense is a proteome-wide tool, its integration in UniProt significantly expands the information on the impact of amino acid changes in the human proteome, and allows users to enrich variants with functional and structural data. AlphaMissense variants and scores are displayed along with UniProt functional sites and AlphaFold 3D structure visualisation in the ProtVista protein sequence viewer.
ProtVar, a tool developed by EMBL-EBI allowing users to contextualise and interpret human missense variation via an unparalleled number of variant types, formats and versions. It combines UniProt functional annotations at a residue level with co-located variants and novel structural predictions of features likely to impact the interpretation of variants. The inclusion of AlphaMissense scores alongside predictions and annotations in ProtVar offers users an even deeper insight into the impact of their variants, linking molecular consequences to human pathogenicity. AlphaMissense scores associated with user entered variants can be viewed and downloaded via the ProtVar website or the ProtVar API.
AlphaFoldDatabase, the resource that provides open access to protein structure predictions from Google DeepMind’s AlphaFold 2 AI system for all known proteins catalogued in the UniProtKB database. Users can now view predicted structures for canonical human proteins and seamlessly toggle between two key visual representations, the quality of structure model (pLDDT) and the average pathogenicity scores from AlphaMissense. These features enable the examination of the missense variations at the residue level in the context of protein 3D structures. The variant data is displayed using an innovative and interactive heatmap and 3D visualisation, offering insights into the broader implications of specific residue changes. The AFDB update significantly aids researchers in identifying structural areas to guide their investigations into protein function and provides easy access to download the data.
Caption: Print screen from AlphaFold Database showing AlphaMissense data integration for Cellular tumor antigen p53.
“The deeper integration of AlphaMissense data into several EMBL-EBI resources is responding to requests from our user communities,” said Maria Martin, Protein Function – Development Team Leader at EMBL-EBI. “AlphaMissense is one more useful resource for scientists working to understand the links between genetic variation and disease, and to investigate which regions of a protein are estimated as more likely to be pathogenic. EMBL-EBI data resources work closely with their user communities to add new features that empower scientists to gain new insights from data, and we’re pleased to expand our offering for exploring genetic variation.”
