During my PhD, I have interpreted rare variants associated with human disease.
Predicting the impact of structural variants on human disease
Genome sequencing has revolutionized the clinical management of rare diseases, and it often leads to diagnoses after standard methods have failed. However, at least half of cases remain undiagnosed. Structural variants (genomic alterations larger than 50 nucleotides, such as deletions and duplications) explain a portion of these undiagnosed cases. We developed a supervised learning method to determine whether a candidate structural variant is likely to cause disease.
Tracking improvements in genetic variant catalogs over time
Researchers have cataloged hundreds of thousands of genetic variants associated with disease. Clinicians use these catalogs to interpret genetic testing results, and the presence of these variants can lead to life-altering medical procedures. Yet, these databases are incomplete and contain errors. We found that misclassified variants were more prevalent in African ancestry individuals. However, databases are improving over time as population-specific allele frequency databases grow.
Machine learning to improve the interpretation of rare variants
Precision medicine promises to tailor treatment to the unique genetic variants in an individual. Yet the majority of missense variants in clinically important genes have uncertain clinical significance. They are also extremely rare. When these variants do appear, clinicians have little to no data to determine a likely disease course or effective treatment. To address these challenges, we propose a genomic learning healthcare system, while considering the technical and ethical challenges.