Hearing loss is caused by some combination of the environment and patient genetics, and is one of the most common sensory deficits. Congenital forms of hearing loss are caused by highly penetrant Mendelian inherited variants in about forty identified genes. The clinical diagnosis is Autosomal Dominant Nonsyndromic Hearing Loss (ADNSHL) when a dominant inheritance pattern is exhibited. Typically, hearing loss is diagnosed with an audiogram, a measure of a patient's hearing levels in each ear at various tone frequencies.
Remarkably, the pattern of ADNSHL patient response to the audiogram depends on the underlying genotype; different genes show different audiogram responses. In this issue, Taylor and colleagues at the University of Iowa and Radboud University Nijmegen Medical Center have developed a novel method that predicts genotype from the audiogram of patients with ADNSHL (Hum Mutat 34:539–545, 2013). Their solution is a multiclass classification machine learning method called AudioGene that encodes the audiograms as features and then predicts the mutated gene. The training data consists of 3,312 audiograms from 1,445 patients. The results are 68% accurate in predicting the causative gene within the top three predictions AudioGene made.
This approach is important and novel because it models the problem of phenotype to genotype prediction (and presumably vice versa) in a unique way. Many Mendelian genetic disorders exhibit incomplete penetrance and variable expressivity. Often, genotype to phenotype correlations remain elusive for all but the most predictive of mutations. Taylor et al. give hope to the community of researchers striving to connect clinical phenotype with molecular causes—a noble effort.