Research On Differential Expression Of RNA-Seq Data

Differential gene and isoform expression analysis is an important research goal in bioinformatics, and receives more and more attention. Differential gene expression analysis mainly compares the genes expression levels under different conditions and identifies the differential expressed genes. It is an important way to understand the gene regulating mechanism. Moreover, detecting differential isoform expression is a feasible method to reveal the change of alternative splicing, which is related to the growth of biological tissues and the occurrence of disease. Therefore, it is of great signifcance to researching the change of alternative splicing.With the fast development of RNA-Seq technology, RNA-Seq has become a standard method for transcriptome analysis, and has been widely applied in detecting differential gene and isoform expression. Compared the “analog” signal in Microarray technology, RNA-Seq experiments produce a large number of reads, which represent “digital” signal to measure expression levels. Therefore, RNASeq has many advantages, such as high throughput, high single-noise ratio and sensitivity etc. However, differential expression analysis in RNA-Seq data still exist some challenges. First, the biases lead to the non-uniform distribution of reads along reference sequences. Next, because of alternative splicing, it is difficult to obtain the read counts for each isoform, and then to detect the differential isoform expression. In order to solve the above problems, we propose two methods---PG_exact test and PG_bayes to detect differential gene and isoform expression. The two methods both base on the results of PGSeq model and adopt various strategies for differential expression analysis. Based on the Negative Binomial of expression distributions in PGSeq model, PG_exact test adopts exact test to detect differential expressed genes and isoforms. In the further research, PG_bayes considers more prior information, and uses Bayes factor for differential expression analysis.We applied PG_exact test and PG_bayes to three human datasets and a mouse dataset, and compared its performance with some popular alternatives. Results show that PG_exact test and PG_bayes perform favorably in sensitivity and specificity at both the gene and isoform level. Due to considering more prior information, PG_bayes shows more significant results.