Title: "Reconsidering Gene Hypothesis Testing in RNA-Seq Experiments"
Speaker: Jeremiah Rounds, Department of Statistics, Purdue University

Place: Lilly (LILY) Hall G126
Date: September 23, 2014; Tuesday
Time: 4:30pm

We examine popular methods for hypothesis testing in next-generation RNA sequencing experiments. A typical analysis of RNA-Seq data proceeds by selecting genome or transcriptome features for counting fragment alignments. Each plausible feature set generates a set of testable hypotheses. We enumerate these testable hypotheses and consider each carefully. An important feature set for testing gene differential expression is generated by the exon-union model (in which fragments are counted as belonging to a gene count if they align within any exon associated with that gene). The exon-union model can be criticized, in that, a gene with multiple mRNA isoforms is not necessarily truly differentially expressed when, due to experimental treatments, the isoforms of its expression change relative expected abundance. We address this criticism by considering a novel new hypothesis test utilizing sub-exon ranges of genome features in conjunction with the exon-union model. We demonstrate, through transcript simulations, that this test, in certain circumstances, has higher true positive rates and lower false discovery rates than just utilizing the exon-union model for gene-wise hypothesis testing. We caution that care must be taken in interpreting these claims, as the null hypothesis being tested has been altered, but that is the point of reconsidering gene hypothesis testing in RNA-Seq.

Associated reading:
1. Trapnell et al. 2013. Differential analysis of gene regulation at transcript resolution with RNA-seq. Nature.

Click here www.stat.purdue.edu/~doerge/BIOINFORM.D/FALL14/sem.html for a full schedule of BIOINFORMATICS SEMINARS, past and present.