Identification of a robust non-coding RNA signature in diagnosing autism spectrum disorder by cross-validation of microarray data from peripheral blood samples

Abstract

Novel molecular signatures are needed to improve the early and accurate diagnosis of autism spectrum disorder (ASD), and indicate physicians to provide timely intervention. This study aimed to identify a robust blood non-coding RNA (ncRNA) signature in diagnosing ASD. One hundred eighty six blood samples in the microarray dataset were randomly divided into the training set (n = 112) and validation set (n = 72). Then, the microarray probe expression profile was re-annotated into the expression profile of 4143 ncRNAs though probe sequence mapping. In the training set, least absolute shrinkage and selection operator (LASSO) penalized generalized linear model was adopted to identify the 20-ncRNA signature, and a diagnostic score was calculated for each sample according to the ncRNA expression levels and the model coefficients. The score demonstrated an excellent diagnostic ability for ASD in the training set (area under receiver operating characteristic curve [AUC] = 0.96), validation set (AUC = 0.97) and the overall (AUC = 0.96). Moreover, the blood samples of 23 ASD patients and 23 age- and gender-matched controls were collected as the external validation set, in which the signature also showed a good diagnostic ability for ASD (AUC = 0.96). In subgroup analysis, the signature was also robust when considering the potential confounders of sex, age, and disease subtypes. In comparison with a 55-gene signature deriving from the same dataset, the ncRNA signature showed an obviously better diagnostic ability (AUC: 0.96 vs 0.68, P