In late September, we will be introducing a significant upgrade to our ACMG Classifier. In line with recent ClinGen guidelines, we will be implementing a points-based score which will reduce the number of variants classified as ‘VUS’, by reducing sensitivity and increasing specificity. This will, initially, be deployed on VarSome.com. A subsequent roll out to VarSome Clinical will take place once we have received and addressed feedback from our users.

Overview

The growing adoption of next generation sequencing (NGS) in clinical laboratories over the past decade resulted in the development of standardized guidelines for the interpretation of genetic data. The American College of Medical Genetics and Genomics (ACMG) developed standard terminology for the classification of variants identified in Mendelian disorders; ‘pathogenic’, ‘likely pathogenic’, ‘uncertain significance’, ‘likely benign’, and ‘benign’. They also recommended a process that would allow clinicians and researchers to assess available evidence to confidently allocate one of these categories to their variants. 

Examining and interrogating the available evidence is a lengthy manual process, involving querying numerous data sources and in silico prediction tools. This bottleneck is what VarSome was built to address. Rather than needing to have a library of tabs open, researchers and clinicians can view 140 data sources in one place giving them an easy to read overview of available evidence for their variant of interest. The VarSome team also developed an algorithm to automatically apply ACMG’s interpretation criteria across the aggregated database to classify variants based on the ACMG terminology.

Since the implementation of VarSome’s ACMG classifier, it has become a ubiquitous tool trusted by research and clinical laboratories around the world to power variant interpretation pipelines.

In the coming weeks we will be releasing a major update to our ACMG classifier. This update implements several recent ClinGen guidelines:

1. PM2: Recommendation for Absence/Rarity Criterion PM2 (Version 1.0)
2. Fitting a naturally scaled point system to the ACMG/AMP variant classification guidelines
3. PP3/BP4: Evidence-based calibration of computational tools for missense variant pathogenicity classification and ClinGen recommendations

The impact of these implementations is that we will have a lower sensitivity and higher specificity. As a result, the overall number of called variants of ‘uncertain significance’ will decrease in favor of an increase in ‘benign’ variants.

Implementation of recent ClinGen guidelines

1. PM2: Recommendation for Absence/Rarity Criterion PM2 (Version 1.0)

The weight of Criterion PM2 (“Absent from controls, or at extremely low frequency if recessive”) will now only ever be triggered with strength “Supporting” instead of “Moderate”. 

2. Fitting a naturally scaled point system to the ACMG/AMP variant classification guidelines

We are implementing the proposed points-based system for evaluating the ACMG verdict, the points represent a log-scale of the odds of a variant being pathogenic (or benign) as follows:

• Supporting = 1 point
• Moderate = 2 points
• Strong = 4 points
• Very Strong = 8 points

A total score is computed as sum(pathogenic evidence) - sum(benign evidence) and the following thresholds applied:

• Pathogenic if >= 10
• Likely Pathogenic if >= 6
• Uncertain if >= 0
• Likely Benign if >= -6
• Benign if <= -7

This is a very straightforward system that is clear and unambiguous. It also makes it possible to measure whether any individual rule and strength is exceeding, or failing to meet, the level of evidence required.

3. PP3/BP4: Evidence-based calibration of computational tools for missense variant pathogenicity classification and ClinGen recommendations

ACMG have proposed an evidence-based method of calibrating all in-silico predictors, comparing them to ClinVar data. We have used this to calibrate over 30 in-silico predictors, giving us an objective measure and rank of the accuracy of individual predictors. As a result, we have selected:

• BayesDel_addAF for missense variants
• CADD (VarSome Premium) for truncating / non-coding variants
• DANN for non-VarSome Premium users
• scSNV for splice-site prediction
• phyloP100Way for conservation

VarSome will now display in-silico predictions using these new scores presenting a graduated response:

Rules PP3 (Multiple lines of computational evidence support a deleterious effect on the gene or gene product) and BP4 (Multiple lines of computational evidence suggest no impact on gene or gene product) will now trigger with varying strengths.

Further information

A more detailed explanation of the changes will be provided in the accompanying release note of VarSome 11.4, which will be released in mid to late September. This update will only apply to the VarSome, VarSome Premium and VarSome API ACMG classification tool through which we will gather feedback before deploying to VarSome Clinical.

If you have any questions or comments regarding the upcoming improvements to the ACMG classifier, we are always happy to hear from you.