Statistical Bioinformatics Webinar: Sharo -- Computational Methods Improve and Refine Clinical Structural Variant Interpretation

Presented by Mr.  Andrew Sharo, University of California, Berkeley 

Computational methods are rapidly improving our ability to predict which germline
variants cause rare Mendelian disease.  The applications are startling.  Consider
Kathleen Folbigg, who is serving a 30-year prison sentence for the alleged murder of her
four children.  Years later, scientists have found that her children inherited rare
variants that may explain their sudden death.  Will variant interpretation eventually
exonerate Kathleen? More commonly, clinical geneticists must identify one or two
disease-causing variants among millions of neutral variants in the genome of an
individual with a rare disease.  However, at least half of these cases remain
unresolved, even after whole genome sequencing.  Structural variants may be the cause of
a portion of these unresolved cases.  We have developed StrVCTVRE, a random forest
method, to prioritize structural variants that overlap exons.  StrVCTVRE will allow
clinicians to eliminate half of structural variants from consideration with 90%
sensitivity.  I will also discuss our analysis of cataloged pathogenic variants, those
variants that have been identified by clinical laboratories or researchers to cause
disease.  We consider two popular databases, ClinVar and HGMD.  Using population
sequencing datasets, we find that pathogenic HGMD variants imply two orders of magnitude
more affected individuals than ClinVar.  We also find that individuals of African
ancestry are five times more likely to be predicted to be affected when HGMD variants
are used.  Encouragingly, more recent clinical variant interpretation recommendations
removed much of the ancestry skew.