Bioinformatic analysis of microarrays:A comparative study of data-processing algorithms

ARIEL CHERNOMORETZ

Orlando, Florida, USA.

Conferencia; II International Conference on Cybernetics and Information Technologies. Bioinformatic Analyses and Applications in Functional Genomics Session.; 2005

Reproducibility of Affymetrix microarray data from in vivo expe <!-- @page { size: 8.5in 11in; margin: 0.79in } P { text-indent: 0.87in; margin-bottom: 0in; line-height: 200%; text-align: justify } P.western { font-family: "Book Antiqua", "Palatino", serif; font-size: 12pt; so-language: en-US; font-weight: medium } P.cjk { font-size: 12pt; font-weight: medium } --> Reproducibility of Affymetrix microarray data from in vivo expe <!-- @page { size: 8.5in 11in; margin: 0.79in } P { text-indent: 0.87in; margin-bottom: 0in; line-height: 200%; text-align: justify } P.western { font-family: "Book Antiqua", "Palatino", serif; font-size: 12pt; so-language: en-US; font-weight: medium } P.cjk { font-size: 12pt; font-weight: medium } --> In this study we present a series of experiments performed to analyse the performance of various data processing algorithms to transform Affymetrix microarray probe raw values into gene expression measurements. According to this, we implemented a robust methodology to detect modulated genes, namely the modified limit fold change (MLFC) model. In our approach, the inherent variability found in within-treatment replicates, is used to identify differentially expressed genes whenever a fold-change is detected outside this reproducibility limits. The respective lists of identified modulated genes were then analyzed in terms of reproducibility of results and validated via quantitative real time PCR assays. The consistency of the lists of modulated genes was studied analyzing the reproducibility of results in triplicate assays. As a general result, the obtained data indicate that biological variability can be reduced in pooled samples. In addition, we found that the most robust lists of genes were produced by the RMA, GCRMA, and MBEI methodologies. At the confidence level considered, all these methods produce very reliable results (~93%) as the validation rates by Q-RTPCR are concerned. The MLFC method turns out to be a well-balanced detection procedure able to identify modulated genes along the entire range of expression intensities for every analyzed preprocessing method. The identified genes had a high QRTPCR-validation rate whatever the preprocessing method used. In addition, the RMA algorithm seemed to be the best choice to process chip data, as it provides the most extensive list of modulated genes, with a high validation rate by Q-RTPCR (94%). ries of experiments performed to analyse the performance of various data processing algorithms to transform Affymetrix microarray probe raw values into gene expression measurements. According to this, we implemented a robust methodology to detect modulated genes, namely the modified limit fold change (MLFC) model. In our approach, the inherent variability found in within-treatment replicates, is used to identify differentially expressed genes whenever a fold-change is detected outside this reproducibility limits. The respective lists of identified modulated genes were then analyzed in terms of reproducibility of results and validated via quantitative real time PCR assays. Results The consistency of the lists of modulated genes was studied analyzing the reproducibility of results in triplicate assays. As a general result, the obtained data indicate that biological variability can be reduced in pooled samples. In addition, we found that the most robust lists of genes were produced by the RMA, GCRMA, and MBEI methodologies. At the confidence level considered, all these methods produce very reliable results (~93%) as the validation rates by Q-RTPCR are concerned. .Conclusions The MLFC method turns out to be a well-balanced detection procedure able to identify modulated genes along the entire range of expression intensities for every analyzed preprocessing method. The identified genes had a high QRTPCR-validation rate whatever the preprocessing method used. In addition, the RMA algorithm seemed to be the best choice to process chip data, as it provides the most extensive list of modulated genes, with a high validation rate by Q-RTPCR (94%).