MLb-LDLr: A Machine Learning Model for Predicting the Pathogenicity of LDLr Missense Variants

Abstract

Untreated familial hypercholesterolemia (FH) leads to atherosclerosis and early cardiovascular disease. Mutations in the low-density lipoprotein receptor (LDLr) gene constitute the major cause of FH, and the high number of mutations already described in the LDLr makes necessary cascade screening or in vitro functional characterization to provide a definitive diagnosis. Implementation of high-predicting capacity software constitutes a valuable approach for assessing pathogenicity of LDLr variants to help in the early diagnosis and management of FH disease. This work provides a reliable machine learning model to accurately predict the pathogenicity of LDLr missense variants with specificity of 92.5% and sensitivity of 91.6%.

Keywords: ANN, artificial neural network; AUROC, area under the receiver operating curve; EGS, expert-guided selection; ESEA, Excel Solver Evolutionary algorithm; FH, familial hypercholesterolemia; LDA, linear discriminant analysis; LDL receptor; LDL, low-density lipoprotein; LDLr, low-density lipoprotein receptor; LNN, linear neural networks; ML, machine learning; MLP, multilayer perceptron; MLb-LDLr, machine-learning–based low-density lipoprotein receptor software; RBF, radial basis function; UTR, untranslated region; familial hypercholesterolemia; machine learning software; pathogenicity; prediction.