Nucleotide composition based measurement bias in high throughput gene expression studies

High throughput gene expression profiling methods suffer from various sources of measurement bias inherent to the experimental procedures used. Most of the commonly used data standardization methods, designed to reduce the sample-tosample variability of technical origin, do not account for probe-or transcript-specific effects. However, the efficiency of RNA isolation, cDNA synthesis and amplification does depend on the percentage of GC nucleotides in the transcript sequences and therefore constitutes a strong bias for the analysis of gene expression data. This work is focused on analysis of how and to what extent GC-content bias of oligonucleotide microarray probes affects the measurement data. We propose amechanism explaining this phenomenon, the implications of GC-content bias for differentially expressed genes (DEGs) detection, and propose a new data standardization method, which by using sample-specific background intensity estimation and LOESS regression, allows to counteract the described effects.