Objectives: To identify molecular signatures in gene expression profiles of peripheral blood from individuals with ASD.
Methods: Blood gene expression profiles of 256 ASD and control samples from Children's Hospital Boston (CHB) and the Translational Genomics Research Institute (TGEN) were analyzed with a machine learning method, Support Vector Machine (SVM). We used a dataset from one institution to build a predictor and tested the other set to validate the accuracy of the predictor.
Results: The comparison of differentially expressed genes from CHB and TGEN revealed a signature of 25 genes (16 up-regulated, 9 down-regulated) in common (nominal p-value < 0.001). Using these 25 genes, we were able to predict ASD with an accuracy of 67% when the predictor was built with the TGEN dataset and tested on the CHB dataset, and 78% when the predictor was from CHB and tested on the TGEN dataset. Interestingly, geneset analysis showed several perturbed pathways such as dendrite morphogenesis.
Conclusions: Blood gene expression profiling predicts ASD with relatively high accuracy, and shows possibility of surrogate biomarker to aid accurate and early diagnosis of ASD as well as providing interesting clues to disease process.