ProtVar update: a discovery engine for genetic missense variation

The human proteome contains millions of missense variants, but only 2% have clinical annotation. For clinical geneticists, protein scientists and drug discovery researchers, the bottleneck is rarely finding variants – it’s working out what they do. 

ProtVar, EMBL-EBI’s resource for interpreting human missense variation, has just undergone a major update. By combining expertly curated knowledge and predictions with AI searching and matching, ProtVar now adds further biological context and clear visualisations that reveal the biological mechanisms behind each variant. Understanding how missense variants affect protein structure and function is vital for identifying drug targets and developing new therapeutics.

Instantly query a single gene or entire database

ProtVar bridges the gaps between protein sequence, structure and function. It integrates consequence predictions with data from UniProt, PDBe, Ensembl, Open Targets, AlphaFold DB and many other sources to provide functional and structural context for interpreting missense variants. 

“In the three years since we launched ProtVar, we have spoken to clinicians, drug developers and researchers to understand how they use the tool,” explains James Stephenson, Lead Data Scientist from the Protein Function Development team at EMBL-EBI. “The first iteration of the resource enabled users to look up, annotate and interpret missense variants from their own experiments. But not all users have their own data; some want to browse the database. To meet that need, we have updated ProtVar to allow retrieval and filtering of annotations covering over 500 million potential variants in the human proteome.”

Filtering by a range of functional and structural fields such as post-translational modifications (PTMs), active sites, binding to ligands, disease association and protein-protein interfaces facilitates the retrieval of the most relevant variants from much larger lists. 

“The new ProtVar release represents a major advancement over the original tool, enabling the community to rapidly retrieve bespoke variant lists unavailable elsewhere,” said Dr Matthew Childs, MRC Human Functional Genomics Cluster Lead.

AI-enabled search

The new search functionality utilises AI to surface the most relevant variants in real time, matching proteome wide annotations to plain-language queries. What once took days of searching across multiple databases now takes seconds, supporting rapid, iterative hypothesis testing. 

Researchers can search using high level concepts to uncover associated variants and synthesise complex biological knowledge. In each case the meaning of the search term is evaluated using one of multiple embedding models to collate related data rather than a key word search. Semantic searches can include disease terms or phenotypes, functional impacts or drug response and pathways. The new search is still in an experimental phase and the team is continually improving and developing the functionality.

“Building on ProtVar’s established role in interpreting clinically relevant variants, this update introduces significant improvements that bring greater granularity and confidence to the data — making interpretation more straightforward and ultimately more impactful for the research community,” said Dr Vanessa Shimizu, Associate Professor in Human Genetics of Infectious Diseases, Imperial College London.

This major update evolves ProtVar from an annotation look up service to a powerful discovery engine enabling researchers to contextualise missense variants with unprecedented flexibility.

Visit the ProtVar website to explore the new tool, and share your feedback with the team to help improve it. 

Please note that artificial intelligence enabled searches may include inaccuracies. It is always recommended to independently verify critical information.