Using Differential Expression Analyses of Transcriptomes to Examine Gene Dysregulation in Autism

Abstract

Recent advances in high-throughput RNA sequencing (RNA-Seq) have made it possible to conduct large-scale analyses of transcriptomes without the use of microarrays. In particular, differential expression analysis can be conducted to determine whether certain genes or exons are significantly expressed amongst multiple samples, provid- ing a more computational perspective to studying the molecular and genetic basis of diseases. The study of autism spectrum disorder, a neurodevelopmental disorder that exhibits a high degree of heterogeneity and has been linked to hundreds of genes, could benefit greatly from differential expression analysis. This paper assesses the use of different normalization and analysis models for conducting differential expression analysis on RNA-Seq data from brain tissue samples obtained from both autism and control patients. Four different analytic models, DESeq2, edgeR, baySeq, and EDASeq normalized linear mixed effects regression, were applied to the gene counts. Surprisingly, each of the methods yielded fairly different sets of differentially expressed genes. However, when gene ontology, pathway association, and disease association analyses were conducted on each set of genes, the specific pathways and functions that were linked to each method were fairly similar, despite each method producing mostly different genes. In addition to many genes and pathways previously linked to autism, such as MAPK9, PAK3, and the ErbB signaling pathway, there were some significant genes and pathways that were not as strongly linked to autism, such as the NOTCH signaling pathway and the EGLN3 gene frequently associated with cancer.