TCGA data, Num_Probes and Segment_Mean

Question

[joshwarrick]

I'm starting to use TCGA data. Level 3 data SNP data determining CNV have the following format:

Sample                                                   Chromosome     Start       End        Num_Probes      Segment_Mean
BEDEW_p_TCGA_FFPE_7_13_N_GenomeWideSNP_6_A05_1347396     1              61735       836713     68              -0.0961
BEDEW_p_TCGA_FFPE_7_13_N_GenomeWideSNP_6_A05_1347396     1              836746      3208375    663             0.4065
BEDEW_p_TCGA_FFPE_7_13_N_GenomeWideSNP_6_A05_1347396     1              3212946     16007300   7302            0.0227
BEDEW_p_TCGA_FFPE_7_13_N_GenomeWideSNP_6_A05_1347396     1              16007754    16088317   33              0.4554

The TCGA consortium includes segments with very small segment means (e.g. 0.02, line 3 above), but a large number of involved probes (e.g. 7302, line 3 above). Does anyone have any opinion on how to interpret these? Such a large number of involved probes makes me think they're more than noise, but the small segment mean suggests there's barely an increase in this segment. Does it represent a subclone comprising a small fraction of the tumor?

Answer 1

[Mattias Aine]

These small variations in segment levels are only there because there is uncertainty in calling the "true" level. There is of course underlying heterogeneity in many tumor samples, but 0.02 is not really even a deviation from 0. If you plot the distribution of discrete levels among all chromosomes for the sample you will see that there is a range around 0 for your normal copy number segments and (ideally) discrete deviations from this representing gains and losses. For subclonality you could of course look for deviations from expected levels, but that can just as well come from adjacent normal cells or infiltrating immune cells. In general I would say its good to look at the distribution of levels for each tumor to get a feel for the data.

[joshwarrick]

Thanks for the reply. That makes a lot of sense.