RESULTS
In order to achieve the goal to prioritize VUS with a higher probability
of being pathogenic and overcome the limitations of current predictor
tools, we developed three models comparing three machine learning
approaches (Random Forest, Support Vector Machine and a Neural Network
with a Five-layer multilayer perceptron architecture). Models were
trained with a set of 82,426 high quality variants from the ClinVar
database and tested with a set of variants that had been classified as
VUS anytime during the last three years, but had been reclassified with
high confidence in any of the 4 informative categories (Pathogenic,
Likely pathogenic, Likely Benign, Benign). To increase the size of the
training set and ease the interpretation of results we merged the
Pathogenic and Likely pathogenic categories into a unique Pathogenic
category, and the Benign and Likely Benign category into a unique Benign
category.