I have recently started to use bioconduct to analyze a time course study with U133A and B, which has 7 time points including time zero (T0). Each time point has three replicates and every one is compared to T0. With R 1.8.1 and affy package 1.3.28, I tested gcrma, rma, Li-Wong algorithm (PM only, subtractMM) and MAS (mas5). After importing the normalized data into GeneSpring, I did two further normalization method for each data set (rma and gcrma data was transferred to linear scale first): (A) per gene normalized to T0 mean only (B) per chip normalization to median followed by per gene to T0, since one GS tech support person told me to do the per-chip normalization in GS and also I was told no further per chip normalization needed from Bioconductor user group. I used the following step to filter the significant changing genes: filter by call flag; by control signal (GeneSpring feature, no effect on rma and gcrma data set); t-test and ANOVA. The results were quite different: from 46 (Li-Wong PMonly (A)) to 232 (MAS(B)). But I did find the following results, for which I need some comments and suggestions. (1) except gcrma and rma, for a given data set, (B) method generated more genes than (A) with a maximum close to 3 times different in MAS; (2) gcrma and Li-Wong subtractMM algorithm generated two to three times more genes than rma and Li-Wong PM only; (3) I did a condition tree (a clustering analysis on time condition, not for gene) with different similarity measurements (standard, smooth, change, pearson, spearman, distance correlation) using their own gene lists and own data for each method. Surprisedly, MAS(B) give the better results: the clustering samples matches with the time course sampling order, followed by MAS(A), gcrma(B), Li-Wong PMonly(B), etc. Generally, (B) give better condition tree than (A) (except Li-Wong subtractMM). Any comments and suggestions are appreciated Lizhe -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/ms-tnef Size: 5134 bytes Desc: not available Url : https://www.stat.math.ethz.ch/pipermail/bioconductor/attachments /20040329/bc0eca2a/attachment.bin