Abstracts

Rationale: Over the past decade, more than 100 genetic etiologies have been identified in developmental and epileptic encephalopathies (DEE). Nevertheless, correlating genetic findings with clinical features at scale has remained a hurdle due to a lack of frameworks and techniques for analyzing sparse heterogenous clinical data. Here we assess whether computational phenotypes can be used in individuals with whole-exome sequencing data to identify relevant similarities in individuals with DEE and shared genetic etiologies. Methods: We used Human Phenotype Ontology (HPO) terms to annotate phenotypic data in 813 individuals with DEE and existing trio whole exome sequencing data. We then assessed observed-versus-expected phenotypic similarity using HPO-based semantic similarity analysis for individuals with de novo variants in the same gene. We developed and tested two different semantic similarity algorithms for these analyses and determined specific HPO terms that drove phenotypic similarities for discrete genetic etiologies. Results: In summary, we analyzed 27,729 HPO terms in 813 individuals with whole exome trio data, including 1,486 unique terms. Of 159 genes with two or more de novo variants, 12 genes had a phenotypic similarity higher than expected by chance including SCN1A (n=15, p<0.001), STXBP1 (n=12, p=0.003), and KCNB1 (n=6, p=0.006). Other genes such as GRIN1 (n=4, p=0.20) and KCNQ2 (n=8, p=0.56) demonstrated no significant similarity captured by our HPO-based methodology. Decomposition of phenotypic similarity revealed gene-specific signatures including “Complex febrile seizures” (HP:0011172; p<0.001) and “Focal clonic seizures” (HP:0002266; p<0.001) in SCN1A, “Absent speech” (HP:0001344; p<0.001) in STXBP1 and “EEG with generalized slow activity” (HP:0011198; p=0.015) in SLC6A1. Conclusions: We demonstrate that HPO-based phenotype similarity analysis captures significant and unique disease profiles revealing the breadth of the phenotypic spectrum in genetic epilepsies. Computational phenotyping can be used to generate statistical evidence for disease causation using phenotyping algorithms analogous to the traditional approach of defining disease entities through similar clinical features. Funding: Children's Hospital of Philadelphia