medRxivGenetic & Genomic MedicinePreprint
Automating the Clinical Interpretation of LDLR, APOB, and PCSK9 Variants: A Web-Based Platform for FH Diagnosis
Helene Gellert‐KristenseniD, Tim Møller EyrichiD, Stefan StenderiD
Abstract
BackgroundClinical interpretation of LDLR, APOB, and PCSK9 variants is essential for diagnosing familial hypercholesterolemia (FH), yet manual evidence curation is labor-intensive and prone to variability. MethodsWe developed a web-based platform (https://fh-interpret.shinyapps.io/beta/) to automate the synthesis of multi-dimensional evidence. The tool integrates population frequencies (gnomAD), phenotype associations (UK Biobank), AI-based pathogenicity scores (AlphaMissense), high-throughput functional data from saturation mutagenesis, and established clinical classifications (ClinVar, HGMD). ResultsThe application streamlines the diagnostic workflow by automatically synthesizing: (1) population frequencies from gnomAD to assess variant rarity; (2) statin-adjusted LDL-cholesterol association analyses in [~]470,000 UK Biobank participants; (3) AlphaMissense pathogenicity predictions; (4) functional scores for [~]17,000 LDLR variants quantifying impact on LDL uptake and LDLR cell-surface abundance; (5) existing clinical evidence from ClinVar and HGMD; and (6) identification of pathogenic variants at identical amino acid positions (ACMG PM5). The tool generates narrative evidence summaries to facilitate standardized ACMG/AMP classification. ConclusionsBy providing real-time access to massive-scale genomic, phenotypic, and functional datasets within a single platform, this tool streamlines the FH diagnostic workflow and supports more consistent and efficient clinical variant interpretation. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=126 SRC="FIGDIR/small/26343831v1_ufig1.gif" ALT="Figure 1"> View larger version (42K): org.highwire.dtl.DTLVardef@a7d316org.highwire.dtl.DTLVardef@1370aborg.highwire.dtl.DTLVardef@49f9faorg.highwire.dtl.DTLVardef@ba936a_HPS_FORMAT_FIGEXP M_FIG C_FIG
- organism: human
- assay: GWAS summary statistics
- assay: whole-exome sequencing
- research focus: variant effect prediction
- research focus: software / tool
- research focus: database / resource
- approach: statistical model
In-depth summary
Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (13 Jan 2026) · CC BY 4.0
A Shiny web platform pulls together gnomAD frequencies, UK Biobank LDL-C associations, AlphaMissense, LDLR deep mutational scanning and ClinVar/HGMD to draft ACMG/AMP evidence for familial hypercholesterolemia variants.
Sequencing has moved the familial hypercholesterolemia (FH) diagnostic bottleneck to interpreting LDLR, APOB and PCSK9 variants, whose ACMG/AMP evidence is spread across gnomAD, ClinVar and the literature and is slow and error-prone to curate by hand. The platform brings these sources together with biobank LDL-C phenotypes and saturation mutagenesis scores to draft standardized evidence summaries in one place.
Key results
ShownOne R/Shiny interface collects per-variant evidence for LDLR, APOB and PCSK9 from population, biobank, GWAS, in silico, functional and clinical sources. It also looks up pathogenic missense variants at the same residue (ACMG PM5).
Sources: gnomAD v4.1.0 (∼1.6 million alleles); UKB statin-adjusted LDL-C regressions in ∼470,000 participants (LDL-C ×1.42 if on statins); GLGC 2017 exome chip GWAS (∼300,000 individuals); AlphaMissense cutoffs <0.34 benign, >0.56 likely pathogenic; deep mutational scan (DMS) of ∼17,000 LDLR variants, <0.5 of wild-type called impaired. · Abstract; Application Development
SuggestedThe combined evidence supports upgrading LDLR p.Leu479Pro from variant of uncertain significance (VUS) to pathogenic. UK Biobank carriers show a large LDL-C elevation, and the in silico and functional data agree.
11 UKB heterozygotes: mean statin-adjusted LDL-C 6.19 vs 3.77 mmol/L in non-carriers (2.42 mmol/L higher; p = 5 × 10^-19); AlphaMissense 0.99; DMS showed impaired LDL uptake and LDLR surface abundance; ClinGen FH expert panel had called it VUS (April 2025). · Case Study 1: Reclassifying LDLR p.Leu479Pro
SuggestedLDLR p.Glu380Lys sits at a residue that carries another pathogenic variant, yet its evidence points to benign. This shows how a PM5-style same-position inference can mislead.
29 UKB heterozygotes: mean LDL-C 3.73 vs 3.77 mmol/L in non-carriers (p = 0.86); AlphaMissense 0.07 (likely benign); functional assays found no loss of LDL uptake or surface abundance; listed as VUS in ClinVar. · Case Study 2: Evidence for Benign Classification…
ShownThe interface lets users explore variants interactively. Selecting a variant updates a genotype-stratified LDL-C plot and a narrative summary that combines all evidence.
Four-part layout: gene selector; variant table with global search by nucleotide or amino acid change and sortable columns (e.g., ranking by p-value); central plot of mean statin-adjusted LDL-C with 95% CI by genotype; narrative text above the plot. · Application Functionality and Case Studies
Methods, data and reproducibility
R/Shiny application that joins per-variant evidence (gnomAD frequency, UK Biobank linear-regression LDL-C effects adjusted for age, sex and ten ancestry PCs, GLGC GWAS effects, AlphaMissense, LDLR deep mutational scanning, ClinVar/HGMD, PM5 same-residue lookup) and outputs a draft ACMG/AMP narrative with genotype-stratified LDL-C plots.
- gnomAD v4.1.0 population allele frequencies used to judge variant rarity · human · n = ∼1.6 million alleles
- UK Biobank LDL-C (multiplied by 1.42 for statin users) regressed on genotype with age, sex and ten ancestry PC covariates; counts under 5 masked · human · n = ∼470,000 participants
- GLGC 2017 exome chip GWAS meta-analysis giving independent per-allele LDL-C effects · human · n = ∼300,000 individuals
- AlphaMissense pathogenicity predictions for LDLR and PCSK9 variants
- 2025 deep mutational scanning (saturation mutagenesis) study scoring LDL uptake and LDLR cell-surface abundance · n = ∼17,000 LDLR variants
- HGMD FH-associated disease-causing (class DM) missense, nonsense and splice variants · human
- ClinVar FH-associated variants extracted for LDLR and PCSK9 · human (clinical variant database)
- Methods & tools
- R (version 4.3.2), Shiny for R (version 1.11.1), gnomAD v4.1.0, UK Biobank, GLGC exome chip GWAS meta-analysis, AlphaMissense, LDLR deep mutational scanning (saturation mutagenesis) functional scores, ClinVar, HGMD, Linear regression (age, sex, ten ancestry PCs), ACMG/AMP variant classification criteria (incl. PM5)
- Software & environment
- R version 4.3.2; Shiny for R version 1.11.1; source code and data released under the MIT License.
- Reproducibility gaps
- Carrier counts for variants seen in fewer than 5 UK Biobank individuals are masked per UKB guidelines; the hosted application is served from a beta URL.
Limitations (4)
- UK Biobank participants are mostly of European ancestry, so the phenotype evidence may not generalize to variants more common in other populations. The authors propose adding diverse cohorts such as All of Us. · stated by the authors
- The generated narratives are preliminary drafts meant to support, not replace, clinical judgment. A qualified professional must assign the final classification in light of the patient's clinical context. · stated by the authors
- Utility is illustrated with only two LDLR case studies. No systematic comparison of drafted classifications, curation time or diagnostic yield against expert-panel or manual curation is reported. · noted in this summary
- Evidence coverage differs by gene: AlphaMissense and ClinVar extraction are described only for LDLR and PCSK9, and the functional scan covers only LDLR, so APOB variants get fewer evidence lines. A single 1.42 statin factor may also misestimate individual LDL-C. · noted in this summary
Read this if you curate LDLR, APOB or PCSK9 variants for FH diagnosis and want one tool that combines UK Biobank LDL-C effects, AlphaMissense and LDLR functional scores into ACMG/AMP evidence drafts.
- Citations
- 0
Reproducibility ledger
Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.
Sign in to file a reproducibility report.
Related papers
- The Clinical Genome Resource (ClinGen) Familial Hypercholesterolemia Variant Curation Expert Panel consensus guidelines for LDLR variant classification
Joana Rita Chora, Michael A. Iacocca, … Mafalda Bourbon·Genetics in MedicinePeer-reviewed·Nov 30, 2021
- Functional analysis of LDLR variants reveals that the novel p.(Ser610Tyr) variant in the YMDD motif drastically impacts LDL uptake
Cris Ortiz López, Neila Rodríguez Roca, … Sonia Rodríguez‐Nóvoa·medRxiv·May 16, 2025
- LDLR variant classification through activity-normalized prime editing screening
Phillip J. Zhou, Minja Velimirovic, … Richard I. Sherwood·CirculationPeer-reviewed·Sep 1, 2026
- Population genomic screening leads to improved lipid management in patients with familial hypercholesterolemia
Matthew E. Levy, Kelly M. Schiabor Barrett, … Catherine A. Hajek·Circulation Genomic and Precision MedicinePeer-reviewed·Nov 5, 2025
- Molecular Screening of Familial Hypercholesterolemia in the Icelandic Population
Greg Kellogg, Bolli Þórsson, … Óscar Puig·bioRxiv·Sep 27, 2018
- Coronary Artery Disease Risk of Familial Hypercholesterolemia Genetic Variants Independent of Historical Cholesterol Exposure
Shoa L. Clarke, Catherine Tcheandjieu, … Themistocles L. Assimes·Circulation Genomic and Precision MedicinePeer-reviewed·Feb 10, 2022