Email updates

Keep up to date with the latest news and content from BMC Bioinformatics and BioMed Central.

Open Access Highly Accessed Open Badges Research article

Different effects of the probe summarization algorithms PLIER and RMA on high-level analysis of Affymetrix exon arrays

Yi Qu, Fei He and Yuchen Chen*

Author Affiliations

National Engineering Center for Biochip at Shanghai, Libing Rd. 151, Shanghai, 201203, China

For all author emails, please log on.

BMC Bioinformatics 2010, 11:211  doi:10.1186/1471-2105-11-211

Published: 28 April 2010



Alternative splicing is an important mechanism that increases protein diversity and functionality in higher eukaryotes. Affymetrix exon arrays are a commercialized platform used to detect alternative splicing on a genome-wide scale. Two probe summarization algorithms, PLIER (Probe Logarithmic Intensity Error) and RMA (Robust Multichip Average), are commonly used to compute gene-level and exon-level expression values. However, a systematic comparison of these two algorithms on their effects on high-level analysis of the arrays has not yet been reported.


In this study, we showed that PLIER summarization led to over-estimation of gene-level expression changes, relative to exon-level expression changes, in two-group comparisons. Consequently, it led to detection of substantially more skipped exons on up-regulated genes, as well as substantially more included (i.e., non-skipped) exons on down-regulated genes. In contrast, this bias was not observed for RMA-summarized data. By using a published human tissue dataset, we compared the tissue-specific expression and splicing detected by Affymetrix exon arrays with those detected based on expressed sequence databases. We found the tendency of PLIER was not supported by the expressed sequence data.


We showed that the tendency of PLIER in detection of alternative splicing is likely caused by a technical bias in the approach, rather than a biological bias. Moreover, we observed abnormal summarization results when using the PLIER algorithm, indicating that mathematical problems, such as numerical instability, may affect PLIER performance.