Tools & data
The Variant Browser
A major challenge facing contemporary genomic medicine is the clinical community’s desire for yes/no answers to the nuanced question of whether a specific genetic variant will produce a meaningful phenotype. Today’s framework classifies variants from likely pathogenic / pathogenic to likely benign / benign, with most stuck as a variant of uncertain significance. At VariantBrowser.org we instead present a data-driven estimate of disease penetrance alongside the raw data, in searchable tables, for interpreting variants in KCNQ1, KCNH2, and SCN5A.
Heuristically, the diagnostic information one learns about a variant from its three-dimensional location, in vitro functional data, and in silico predictors is roughly equivalent to clinically phenotyping 10–20 heterozygotes. Published in PLOS Genetics (2020), Circ. Genomic & Precision Medicine (2021), and Genetics in Medicine (2022).
Open-source toolkit
Our methods are released as reusable, documented code. Everything below is on GitHub.
LLM-driven discovery and extraction of per-variant carrier and phenotype evidence from the PubMed literature.
LLM-powered structured extraction of gene, variant, and phenotype data from the published record.
Aggregates AlphaMissense, REVEL, CADD, ClinVar, and gnomAD into one unified, queryable SQLite database.
Deep mutational scanning of KCNH2/Kv11.1 trafficking — perturbation data for thousands of variants at once.
Predicting Brugada syndrome penetrance from NaV1.5 (SCN5A) functional, structural, and sequence features.
Predicting functional perturbation in KV7.1 (KCNQ1) and NaV1.5 (SCN5A) from structure and sequence.
Carrier data and disease-penetrance estimates for CPVT-associated RyR2 variants.
Patient- and variant-specific features for the risk of severe cardiac events in long QT syndrome.
Additional components — the Bayesian penetrance estimator, AlphaFold structural-proximity analysis, and the ClinGen gene–disease curation app — are in active development and available on request.