Application of Artificial Neural Network for Prediction of Risk of Multiple Sclerosis Based on Single Nucleotide Polymorphism Genotypes

The artificial neural network (ANN) is a sort of machine learning method which has been used in determination of risk of human disorders. In the current investigation, we have created an ANN and trained it based on the genetic data of 401 multiple sclerosis (MS) patients and 390 healthy subjects. Single nucleotide polymorphisms (SNPs) within ANRIL (rs1333045, rs1333048, rs4977574 and rs10757278), EVI5 (rs6680578, rs10735781 and rs11810217), ACE (rs4359 and rs1799752), MALAT1 (rs619586 and rs3200401), GAS5 (rs2067079 and rs6790), H19 (rs2839698 and rs217727), NINJ2 (rs11833579 and rs3809263), GRM7 (rs6782011 and rs779867), VLA4 (rs1143676), CBLB (rs12487066) and VEGFA (rs3025039 and rs2071559) had been genotyped in all individuals. We used “Keras” package for generation and training the ANN model. The k-folds cross-validation strategy was applied to confirm model generalization and overfit prevention. The locally interpretable model-agnostic explanation (LIME) was applied to explain model predictions at the data sample level. The TT genotype of the GAS5 rs2067079 had the most protective effect against MS, while the DD genotype of the ACE rs1799752 was the most prominent risk genotype. The accuracy, sensitivity and specificity values of the model were 64.73%, 64.95% and 64.49% respectively. The ROC AUC value was 69.67%. The current study is a preliminary study to appraise the application of ANN for prediction of risk of MS based on genomic data.