Hi. On Fri, Dec 19, 2008 at 4:35 PM, ronmaster <rene.para...@gmail.com> wrote: > > Hi Henrik, > > My colleague approach me with this brillant question relating > normalization on tumor samples. We can read > in an Affymetrix note (Genotyping console manual p173): > > "Probe-level Normalization for Reference Model File Creation > Quantile normalization is recommended for copy number analysis of > association and > cytogenetics samples. Quantile normalization is most appropriate for > samples where most of the > chromosomes are relatively normal. > ******************************************* > IN CONTRAST, many cancer samples contain significant abnormalities > that impact much of the > genome; therefore, median normalization is recommended." > ******************************************* > > We can also read at the point 3.2.2 of your article: "Estimation and > assessment of raw copy numbers at single locus level" that > normalization is undergone with the assumption of a normal distributed > DNS targets across the samples. > > > How do you deal with this type tumor samples?
In CRMA and CRMAv2, we perform a few different calibration and normalization methods: Allele-crosstalk calibration (ACC): This single-array method estimates and correct for the crosstalk between alleles as well as the global offset in data. These estimates are robust against CN aberrations, because they are based on the homozygote SNPs (A, AA, AAA, AAAA, ..., B, BB, BBB, BBBB, ...), but much less so on heterozygote SNPs (AB) and all other non-homozygote SNPs (AAB, AAAB, ABB, ...). In ACC, the signals are rescaled such that their robust average is the same across arrays (an arbitrary constant). In addition, there is a rescaling step, which corresponds to "median normalization" step Affymetrix refers to. This step can be affected if the sample has a large number of aberrations (theoretically only if >50%), which is rare, and if it occurs, it is often not an issue because it is only a global scale factor and any CN changepoint will still remain. Nucleotide-position normalization (BPN; CRMAv2): This single-array method estimates and correct for probe-sequence effects. This method is not robustified and may be affected if there are a lot of aberrations. There is nothing preventing us from robustifying the estimator, say by an iterative reweighed approach), but such an implementation is not in place. We have applied BPN to tumor samples with many aberrations and the results still look good. If you are uncomfortable with this, you may skip the BPN step. Probe summarization (PLM): These various summarization methods - single- or multi-array ones - are all robustified. For the multi-array PLMs, the assumption is that the robust average of all arrays is copy-neutral at any given locus. This assumption is done by Affymetrix' (and dChip's etc.) methods as well. We also do a post-summarization normalization that controls for PCR fragment-length effects. This is also a single-array method that is robust for CN aberrations. As above, it will not break down until there are a great number of aberrations (>50%). Thus, there exist no magic normalization method that deals with extreme number of CN aberrations, and especially not when they are shared by the majority of samples. In the worst case scenario all tumors have the same aberrations and there are no normals; then it is not possible to identify CN aberrations base only on total CN estimates (allele-specific estimates can help, but that's a different story). To summarize, we are aware of the problem you are bringing up and we do design our methods to deal with such cases as good as possible. Cheers Henril > > > cheers > > Rene > > > --~--~---------~--~----~------------~-------~--~----~ When reporting problems on aroma.affymetrix, make sure 1) to run the latest version of the package, 2) to report the output of sessionInfo() and traceback(), and 3) to post a complete code example. You received this message because you are subscribed to the Google Groups "aroma.affymetrix" group. To post to this group, send email to aroma-affymetrix@googlegroups.com To unsubscribe from this group, send email to aroma-affymetrix-unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/aroma-affymetrix?hl=en -~----------~----~----~----~------~----~------~--~---
