Impact of normalization and filtering on linkage analysis of gene expression data

Joseph Beyene,Elena Parkhomenko,Pingzhao Hu,David Tritchler

doi:10.1186/1753-6561-1-s1-s150

Abstract

Using the Problem 1 data set made available for Genetic Analysis Workshop 15, we assessed sensitivity of linkage results to a correlation-based feature extraction method as well as to different normalization procedures applied to the raw Affymetrix gene expression microarray data. The impact of these procedures on heritability estimates and on expression quantitative trait loci are investigated. The filtering algorithm we propose in this paper ranks genes based on the total absolute correlation of each gene with all other genes on the array and has the potential to extract features that may play role in functional pathways and gene networks. Our results showed that the normalization and filtering algorithms can have a profound influence on genetic analysis of gene expression data.

Highlights

Several recent studies have applied traditional quantitative trait linkage analysis to genome-wide gene expression data and have investigated the role of genetic variation in transcription [1]
5.3 the three normalization approaches described earlier (RMA, Gene Chip RMA (GCRMA), and probe logarithmic error intensity estimate (PLIER)) along with our correlationbased feature extraction algorithm in comparison with the probe sets selected by Morley et al [1] following their use of the Affymetrix Microarray Analysis Suite (MAS) algorithm and a variance-based filtering varied from 10% to 43%
We applied three different normalization methods to the Problem 1 data set provided for Genetic Analysis Workshop 15 (GAW15) and carried out expression quantitative trait linkage analysis for 3554 traits

Summary

Introduction

Several recent studies have applied traditional quantitative trait linkage analysis to genome-wide gene expression data and have investigated the role of genetic variation in transcription [1]. These types of studies have the potential to uncover complicated transcriptional control. Gene expression measurements are not generated in a uniform platform and, depending on the particular technology, most high-level statistical analyses using these data are preceded by a number of low-level pre-processing steps. Affymetrix GeneChip arrays have become a widely used microarray platform [2] and there are various algorithms for performing feature extraction and normalization on these high-density oligonucleotide gene expression arrays [3,4]. Findings could be sensitive to variations in (page number not for citation purposes)

Methods

Results

Conclusion