The Use of Machine Learning and Phonetic Endophenotypes to Discover Genetic Variants Associated with Speech Sound Disorder

Abstract

Thirty-four (34) children with reported speech sound disorders (SSD) were recruited for a prior study, as well as 31 of their siblings, many of whom also showed SSD. Using dataclustering techniques, we assigned each child to one or more endophenotypes defined by the number and type of speech errors made on the GFTA-2. The genetic samples of 53 of the participants underwent whole exome sequencing. Variant alleles were detected, filtered, and annotated from the sequences, and the data were filtered using quality checks, annotations, and phenotypes. We then used Random Forest classification to search for associations between variants and endophenotypes. In this preliminary report, we highlight one promising association with a common variant of COMT, a dopamine metabolizer in the brain.