High-Dimensional Feature Characterization of Single Nucleotide Variants in Hypertrophic Cardiomyopathy.

Hypertrophic cardiomyopathy is a genetic disorder that affects the structure of the heart muscle, which can lead to sudden cardiac arrest. The genetic characterization of biomarkers remains an open area, and machine learning techniques are being proposed for its detection. This research aims to apply several of these methods to obtain single nucleotide variants (SNVs). We followed a three-stage approach: First, the initial set of 118142 SNV features were filtered with the union of Manhattan threshold from biostatistics together with the Chi-squared test and with a logistic regression based univariate filtering method, yielding a preselected set of 1974 features; second, linear classifiers (support vector machine and Fisher linear discriminant analysis) identified and ranked the relevant features to distinguish between normal subjects and diseased patients. Finally, two additional techniques (informative variable identifier and Bayesian networks) were used to scrutinize the inter-feature relationships of the SNVs. The results showed a consensus between linear classifiers in which variants with higher weights coincide. The 100 variants with higher weights were visualized to analyze their relationships. To validate the result, the top-ranked variants were checked in the literature. Most of them were directly implicated with the disease or participated in cardiac re-modeling, meaning that these variants can be considered modulators of the disease.