## Protocol 54 Data processing

The image files were imported into the Microarray Suite image analysis software (Version 2.0), which runs as an extension of IPLab Spectrum Software (Scanalytics, Fairfax, VA). The software determined the raw spot intensities of Cy3 and Cy5 and performed a local background subtraction. Empty spots and spots carrying plant sequences were excluded from further analysis. Each dye swap experiment was normalized by applying variance stabilization (see also Chapter 17) (2) using the vsn package of bioconductor (http://www.bioconductor.org). Means of normalized log-products and log-ratios of each dye swap experiment pair were used for further analysis. Normalization procedures were performed using R (http://cran.R-project.org).

To determine a meaningful cut-off value that designates differentially expressed genes we applied a statistical analysis which considers the small number of biological replicates and aims at minimizing the percentage of false positives. To do so, a variance estimation using a pooled estimate of the variance over all genes of three self-to-self comparisons with RNA from undifferentiated cells was performed. A similar approach has been described by Sabatti et al. (3). In order to use a robust variance measurement we determined the median of absolute deviation (MAD) as variance estimator. The MAD in all three independent experiments of self-to-self comparison was very similar: 0.297 ± 0.021. Based on this analysis, we can assume a rate of 2% false positives when applying a universal threshold of 2.17-fold change. A rate of 5% false positives can be assumed when applying a threshold of 1.8-fold. We therefore considered all clones above a 2.0-fold change as relevant. Thus, we can assume a false positive rate of 2-5% at this point of analysis.

To extract clones that are of interest for further analysis we concentrated on all clones, whose expression changed more than twofold in at least one of the three time points of differentiation in both experimental series. These were 722 clones in the BDNF series and 624 in the NT4 series. Since both neurotrophic growth factors target the TrkB receptor, they cause very similar molecular effects in the cells. Their functions in vivo, however, do not perfectly overlap (4). To find common biological effects related to neural differentiation we then took the intersection of both lists, amounting to 454 clones. In addition this step further reduced the number of false positives.

## Post a comment