Samuel, Thanks for the note. Regarding your request I would recommend you start with the bioconductor mailing list for help: bioconductor Mailing List <bioconductor at="" stat.math.ethz.ch=""> Best wishes, Rafael Samuel Wuest wrote: > Dear Rafa, > > Hope you are fine? I have become a big fan of you when I came across the > analysis of Affy-Chips, so thanks for the nice toolboxes you have > created so far. > > Although I am not a statistican but a geneticist, I have been chewing > for quite a while on the problem of analyzing chips from amplified RNA, > and couldn't come up with a solution for making an accurate > Present-Absent decision matrix so far? > > I am trying to modify the Affymetrix MAS5 present-absent calls > algorithm, adjusting for probe-position bias in the data from amplified > samples. I have tried the normal MAS5 and the half-price method, but > both of these methods do not cope too well with the amplification bias? > > Up to date, I have the basic ideas and the tools for their > implementation/checking the new algorithm, but would need some help on > certain issues (details, if interested, see below). I think that your > group has the best expertise on the field: do you know anyone that I > could contact for a small collaboration? > > Thanks for any help on this, best regards. > > Samuel > > ------------------------------------------------------ > Wuest Samuel > Smurfit Institute of Genetics > Trinity College Dublin > Dublin 2, Ireland > Phone: +353-1-896 2444 > Mobile: + 353-85-735 5821 > Email: wuests at tcd.ie <mailto:wuests at="" tcd.ie=""> > ------------------------------------------------------ > > Details: (supporting illustration are available if necessary: quality > control images, 3' positional bias, sensitivity index density, ROC > curves of the MAS5 algorithm on my data, etc). > > I am interested in the sexual reproduction of plants, and thus have > isolated single cells of reproductive importance in Arabidopsis, such as > the egg cell. The samples needed two round of RNA amplification, and > there was a slight problem with RNA degradation, so it took me quite a > while to get data of suitable quality? > > My goal is the creation of a modified MAS5-calls algorithm that takes > into account both probe sensitivity and amplification bias: > for this, I want to make use of probe-sensitivity indices modelled > across multiple chips, all from samples that have been amplified. This > allows to estimate both, the probe sensitivity and the amplification > bias at the same time. > > The sensitivity-index can be taken as a measure of how reliable a signal > from a certain probe actually is. > So I would change the one-sample Wilcox implemented in the mas5calls > into a two sample Wilcox, comparing the measured signals against a > null-distribution (consisting of 11 zero-values, one for each probe). > For each single probe, an assigned null-value is used for comparison, > and this null-value is chosen according to the modelled > probe-sensitivity ( e.g. based on quantile information of the respective > sensitivity). The new algorithm is assessed using ROC curves that make > use of the knowledge that is available so far for the different cell > types (true positives) and of a negative probes (where reannotation has > led to probes with no targets). > > I would need help on the following points: a) how to create a > null-distribution from which to choose null-values and b) general > feedback on the method (e.g. is there any major statistical pitfalls in > this?). > >