Ensembl release 116 brings updated predictions to help interpret the potential impact of human genetic variants. By introducing two machine learning models and improving existing scores, this update makes it easier to explore and understand variant effects across the website.
The Ensembl website displays a range of computational predictions to help assess the potential pathogenicity of genetic variants. In release 116, we have updated our variant pages by replacing the older MutationAssessor and MetaLR scores with the state-of-the-art AlphaMissense and ESM1b predictors. These annotations are sourced from the dbNSFP v5.2c release.
In addition, this release includes a fix to the way REVEL scores are displayed, ensuring that transcript-specific scores are now reported correctly.
AlphaMissense and ESM1b Scores
AlphaMissense is a deep learning model developed by Google DeepMind that predicts the likely pathogenicity of missense single nucleotide variants (SNVs). It classifies variants as likely benign, pathogenic, or ambiguous, and assigns a score between 0 and 1, with higher scores indicating a greater likelihood of pathogenicity.
ESM1b is a protein language model containing 650 million parameters that predicts the functional impact of coding variants directly from protein sequence information. Like AlphaMissense, it provides a quantitative estimate of variant pathogenicity that can aid variant interpretation. The scores shown in the website can range from -24.358 to 6.937. The smaller the score the more likely the variant is pathogenic.
Together, these predictors represent the latest generation of machine learning approaches for prediction of variant impact.
Where You Can See the New Scores
The new pathogenicity predictions are available throughout the Ensembl website, including:
- Variation Summary page
- Variation Table on transcript pages
- Protein Variation view on transcript pages
Variation Summary Page
On the Variation Summary page, navigate to the Genes and regulation section. The transcript table now displays AlphaMissense and ESM1b scores alongside existing predictors such as SIFT and PolyPhen.
As with other pathogenicity predictors, scores are displayed to three decimal places and colour-coded according to their prediction class. Green indicates lower-risk classifications (for example, tolerated or benign), while red indicates higher-risk classifications (for example, deleterious or pathogenic).
Variation Table on transcript pages
The same pathogenicity scores are also available in the Variation Table on transcript pages, allowing you to view predicted variant impact across the transcript.
Protein Variation view on transcript pages
You can also view AlphaMissense and ESM1b predictions within the Protein Variation view on transcript pages, providing additional context when examining amino acid changes and protein-level consequences.
Update to REVEL Scores
This release also includes a correction to the display of REVEL scores.
Previously, due to a bug in the processing of dbNSFP data, variants with multiple transcript-specific values were not handled correctly. As a result, the REVEL score associated with the first transcript was displayed for all transcripts of a variant.
This issue has now been fixed, and transcript-specific REVEL scores are displayed correctly throughout the website.
For example, for variant rs699 shown in Figure (a), the archived site reports a REVEL score of 0 for transcript ENST00000366667.6. In release 116, the corrected transcript-specific score of 0.157 is displayed.
This improvement ensures more accurate representation of transcript-level pathogenicity predictions and better consistency with the underlying dbNSFP annotations.
Disclaimer
The AlphaMissense Database and any related information provided on or linked from this site are intended for research and theoretical modelling purposes only. Appropriate caution should be exercised when using these data.
The information is provided “as is” without warranty of any kind, whether express or implied. For clarity, no warranty is given that use of the information will not infringe the rights of any third party (and this disclaimer takes precedence over any contrary provisions in the Google Cloud Platform Terms of Service).
The information provided is not intended to be a substitute for professional medical advice, diagnosis, or treatment, and does not constitute medical or other professional advice.
Author: Syed Nakib Hossain
Editor: Sarah Hunt