An embedded and charset-unspecified text was scrubbed... Name: not available Url: https://stat.ethz.ch/pipermail/bioconductor/attachments/20070607/ e78fe830/attachment.pl

We are experiencing a similar memory problem with Illumina SNP chips. We are looking for a general solution using databases to handel very large data sets. Does that make sense? -Simon Date: Thu, 7 Jun 2007 11:43:51 +0200 From: Olivier Nu?ez <nunez@est-econ.uc3m.es> Subject: [BioC] Memory issue in 500k chip analysis To: Benilton Carvalho <bcarvalh at="" jhsph.edu=""> Cc: bioconductor at stat.math.ethz.ch Message-ID: <cfdba356-5927-4f2b-93ae-47dd2f195c8a at="" est-econ.uc3m.es=""> Content-Type: text/plain Dear Benilton, an obvious and possibly naive way to avoid the common memory issues in the analysis of 500K chips with oligo, would be to select a sample of SNPs (likely under a stratified random design) from the array before using snprma. Therefore, my first question is whether or not such a sampling method is feasible under oligo. The "affxparser" library allows to select subset of SNPs but I ignore how oligo-package deals with values of a command like readCelUnits {affxparser}. My second question is about the reliability of such a method: The linear fit you use to perform the normalization is based on the whole set of SNPs and the information from the mapping package. Do you think the preprocessing oligo method remains valid (at least for the selected sample) if it is based only on a sample of SNPs. Thanks for your help. Best. Olivier -- ______________________________________________________________________ __ Olivier G. Nu?ez nunez at est-econ.uc3m.es Universidad Carlos III de Madrid [ Tel : +34 663 03 69 09 ] Departamento de Estadistica [ Fax : +34 91 624 98 49 ] Facultad de Ciencias Sociales C/ Madrid 126 GETAFE 28903 SPAIN Web Page: http://www.est-econ.uc3m.es/onunez