Advanced variant classification framework reduces the false positive rate of predicted loss of function (pLoF) variants in population sequencing data

Abstract

Predicted loss of function (pLoF) variants are highly deleterious and play an important role in disease biology, but many of these variants may not actually result in loss-of-function. Here we present a framework that advances interpretation of pLoF variants in research and clinical settings by considering three categories of LoF evasion: (1) predicted rescue by secondary sequence properties, (2) uncertain biological relevance, and (3) potential technical artifacts. We also provide recommendations on adjustments to ACMG/AMP guidelines's PVS1 criterion. Applying this framework to all high-confidence pLoF variants in 22 autosomal recessive disease-genes from the Genome Aggregation Database (gnomAD, v2.1.1) revealed predicted LoF evasion or potential artifacts in 27.3% (304/1,113) of variants. The major reasons were location in the last exon, in a homopolymer repeat, in low per-base expression (pext) score regions, or the presence of cryptic splice rescues. Variants predicted to be potential artifacts or to evade LoF were enriched for ClinVar benign variants. PVS1 was downgraded in 99.4% (162/163) of LoF evading variants assessed, with 17.2% (28/163) downgraded as a result of our framework, adding to previous guidelines. Variant pathogenicity was affected (mostly from likely pathogenic to VUS) in 20 (71.4%) of these 28 variants. This framework guides assessment of pLoF variants beyond standard annotation pipelines, and substantially reduces false positive rates, which is key to ensure accurate LoF variant prediction in both a research and clinical setting.

Comparison of the ABC and ACMG systems for variant classification

“…Finally, our results suggest that elaborating the ACMG criteria even further, as planned, may not increase precision, as intended, because the selection of criteria in challenging cases (like the ones chosen for this study) appears to be quite subjective (Table 2 and Supplementary File S3), and in clear-cut cases there is little need for additional criteria (see Table 2, where the degree of criteria consensus diminish with clinical complexity). However, the integration of molecular rules or AI-based tools into the classification system, as suggested for loss-of-function [15], splice [7] and missense [16] variants in relation to ACMG classification, is a good idea that can also be most helpful for in the functional step A of the ABC system. Another suggestion has been to add "predisposing" and "likely predisposing" as two new ACMG categories [17], but the basic problem is that one-dimensional ACMG classification will always struggle to categorize the spectrum of causes leading to mendelian conditions into a likelihood-of-pathogenicity framework [18].…”

Exploring penetrance of clinically relevant variants in over 800,000 humans from the Genome Aggregation Database

“…We performed deep case-by-case assessments on the 734 high-confidence pLoF variants found in gnomAD genomes, building on a framework for pLoF variant assessment 25 using conservative rules (Table S5) to exclude variants likely to not result in LoF (Table S6).…”

“…The copyright holder for this preprint this version posted June 13, 2024. ; https://doi.org/10.1101/2024.06.12.593113 doi: bioRxiv preprint 14 After assessment using the LoF classification framework 25 , 137 variants in 236 individuals remained without an explanation. We noted a project-specific enrichment with 25.5% of variants (n=35) or 17.4% of samples (n=41), originating from the 1000 Genomes Project, although the 1000 Genomes Project constitutes less than 5% of gnomAD genome samples.…”

“…The Genome Aggregation Database (gnomAD) is a widely used publicly available collection of population variant data, currently sharing harmonized data from 807,162 individuals, including 76,215 genomes and 730,947 exomes (version 4, released November 2023). The gnomAD dataset has played a key role in supporting the discovery of genes and variants association with genetic diseases, enabling improved variant classification in clinical as well as mechanistic-focused interpretation of variants in disease-discovery research settings [19][20][21] . Investigation of well-established and predicted disease-associated variants in unaffected individuals in gnomAD presents an unexplored opportunity to improve our variant interpretation abilities, increase understanding of variant effect, and inform on mechanisms affecting the penetrance of disease.…”

AutoGVP: a dockerized workflow integrating ClinVar and InterVar germline sequence variant classification

“…Adjusting PVS1 strength is crucial since InterVar can over-interpret loss-of-function (LOF) variants (Singer-Berk et al ., 2023) as it classifies all LOF variants that are in LOF-intolerant genes curated by InterVar as PVS1 (Li and Wang, 2017). AutoGVP can classify any number of variants in standard VCF format.…”