Gene Ontology Analysis for RNA-seq: Accounting for Selection Bias in Differential Expression Detection
Overview
This paper validates the GOseq method as a superior approach to standard Gene Ontology analysis for RNA-seq data. It demonstrates that traditional hypergeometric distribution methods produce biased results because they fail to account for length bias and read count bias in RNA-seq data, whereas GOseq corrects for these technical artifacts to produce more reliable and biologically consistent findings.
What this coursework covers
- Introduction: Establishes the need for improved GO analysis methods that account for RNA-seq technical features
- Key Findings: Explains how selection bias violates hypergeometric distribution assumptions in standard methods
- Results: Demonstrates that GOseq produces higher overlap with microarray data and reveals significant differences when correcting for read count bias
Preview
Student’s Name:
Date:
Instructor:
RNA-seq Data Analysis
The publication is intended to validate the need for new methods of analyzing Gene Ontology for RNA-seq data, and mainly focuses on the GOseq method. GO analysis is generally used in minimizing complexity and highlighting biological processes in genome-wide expression studies. It provides unprecedented detail regarding the transcriptional landscape of organisms. However, standard analysis methods tend to give biased results. As such, there is need for developing techniques that incorporate the technical features of the RNA-seq output.
The published data sets comprise of over 17 million short cDNA reads acquired for both the untreated and treated cell line and run on Illumina’s 1G genome analyzer. The reads were mapped again back to the reference genome, and a recording of the number of reads per gene was taken.
Key findings
Standard methods for examining over-representation of a GO category presume that in the null hypothesis, every gene has equal probability of being perceived to be DE. As such, the number of genes linked to a category that overlap with DE genes set follows a hypergeometric distribution. The existence of selection bias makes genes with more reads to be detected as DE, and, this violates the assumption behind the hypegeometric distribution. For this reason, such standard test distributions ought not to be used for GO analysis with RNA-seq data. New tests with the capability of correcting the selection bias are needed.