Quantitative Biology Colloquium

Variants in the SCN8A gene which codes for the voltage gated sodium channel NaV1.6 cause a wide spectrum of pediatric neurological diseases. With a low occurrence rate and the recent discovery of the SCN8A gene, characterizing the SCN8A Channelopathy provides many challenges. To aid in identifying genotype-phenotype and phenotype-phenotype correlations, a International SCN8A Patient Registry was constructed to collect information on patient’s entire life history. Using the data collected in the Registry, it has been shown that machine learning techniques can differentiate between loss- and gain-of-function variants using basic phenotypic data with 95% accuracy. In an effort to further apply machine learning techniques to characterizing the SCN8A Channelopathy, a series of ordinal logistic regression models have been constructed to classify patients possessing a gain-of-function variant as Mild, Moderate, or Severe. The work presented here examines two approaches applied to a series of three classification models and the synthesis of these six different models to provide clinicians with the best tool to accurately provide a confident prognosis for patients affected by SCN8A Channelopathy. Finally, a machine learning pipeline is proposed for providing an accurate diagnosis, prognosis, and optimal treatment plan for clinicians to provide personalized medicine based off of genotypic and phenotypic data available early in the disease progression.

Place:              Math, 402