High importance
Aug 26, 2026
To evaluate the prevalence of hereditary haemorrhagic telangiectasia (HHT) using genetic data and machine learning techniques, compared to traditional clinical assessments.
The study employed evolutionary history analysis alongside protein-focused machine learning to analyze genetic data related to HHT-associated genes including ALK1, endoglin, and SMAD4. This approach was used to reassess the prevalence rates of HHT and its associated arteriovenous malformations.
Findings indicate that HHT may be 2-12 times more prevalent than previously recognized, suggesting it exceeds the threshold for 'rare diseases.' This could significantly alter the clinical management strategies for asymptomatic individuals identified through genetic testing.
The research may be limited by the availability of comprehensive genetic datasets and the potential for variability in genetic expression, which could affect prevalence estimates. Additionally, the reliance on machine learning methods necessitates cautious interpretation of results in the clinical context.
Understanding the true prevalence of HHT is crucial for redefining diagnostic criteria, improving patient management protocols, and addressing the needs for surveillance and treatment of at-risk individuals.
BACKGROUND: Recent genetic data suggest hereditary haemorrhagic telangiectasia (HHT) is 2-12 times more common than the clinically-ascertained prevalence, potentially above the 'rare disease' designation threshold, and undermining clinical predictions for asymptomatic individuals diagnosed by genetic testing. AIM: To test, we examined if missense variants in HHT disease-causing genes may have been misclassified as pathogenic (LP/P) or benign (B/LB). DESIGN: Evaluation of ClinVar-annotated missense variants in ENG, ACVRL1 and SMAD4. METHODS: Human-independent methods using CodeXome for pan-primate evolutionary history, and AlphaMissense which incorporates AlphaFold predictions for protein misfolding were used to validate/reclassify pathogenic and benign missense variants. RESULTS: ClinVar annotations were commonly conservative with 35-90% of rare missense substitutions in ENG, ACVRL1 and SMAD4 classified as variants of uncertain significance (VUS). CodeXome identified 92% of ClinVar-annotated B/LB variants were shared with other primate species, supporting their benign classification. AlphaMissense metrics strongly correlated with CodeXome, and 380/403 (94.3%) variants matched ClinVar benign-pathogenic annotations. However, a small number of variants showed conflicting classifications with ClinVar: 19/293 (6.5%) appeared to be over-called as LP/P by ClinVar representing 15/408 (3.7%) of genotyped families at Imperial, while 4/110 (3.6%) were apparently under-called as B/LB, and not accessible through clinical gene test reports. Newer pathobiological understanding of variants, and recognition of shared familial tendencies reflecting non-HHT heritable burdens were identified as possible explanations of over-calls. CONCLUSIONS: Our findings suggest tools to simplify variant pathogenicity predictions; molecular diagnoses to revisit for HHT families, but do not materially influence prevalence estimates for 'genetic' HHT, challenging current clinical policies, training and standards.