Planned Analysis: Copy number plot showing recurrently amplified/deleted regions in different PBTA cancers

[PichaiRaman]

The idea would be to visualize trends of deletions / amplifications in each of the cancer types (where relevant). Pointing out highly amplified/deleted regions with oncogenes/tumor suppressors may be interesting.

[cgreene]

@jharenza : is this still blocked? It seems like something that could be done now and if the CNV calls get improved, then this would just get re-run.

[jharenza]

@jharenza : is this still blocked? It seems like something that could be done now and if the CNV calls get improved, then this would just get re-run.

I think I may have added it that way thinking consensus calls @gonzolgarcia creates would go into this, but perhaps consensus for CNV and SV would be separate issues?

[cgreene]

Ok!

[cbethell]

I have begun to tackle this issue. Thus far, I believe the package GenVisR would be useful for this issue based on the literature I found here. If there are any suggestions for the implementation of an alternative package, please feel free to express them here.

[jaclyn-taroni]

Related: #128

[jharenza]

Hi @cbethell ! Thanks for working on this! I also just created an issue #128 for creation of consensus calls for CN to improve our accuracy for downstream analytics like this issue. I imagine the output would still be a SEG file, such that you can start with any of the SEGs we provide. I previously have used GISTIC, but it does not work well for small cohorts (ie you need a large N to call recurrence within a caner). We may want to focus this issue on histologies with a specified >= N so that we have enough power to detect recurrent changes already known in the literature. For example, we have a large cohort of low-grade gliomas, we have a good number of medulloblastomas (could see how many of each subtype we have since we know each subtype has different CNAs - Figure 2), high-grade gliomas. I found this recent paper describing TAGCNA, which may work similar to GISTIC.

[jaclyn-taroni]

Summarizing my understanding of the status of this issue:

• The mention of GISTIC and TAGCNA = I understand we're looking for significant amplifications or deletions and therefore some metric and test for significance.

• As far as visualization is concerned, @jharenza has requested

  having this plotted in GISTIC-type format for the different histologies or as an IGV-type figure, grouped by histology, would be most effective for showing recapitulation of previous findings. Addition of R script to compare CNV caller output #142 (comment)

  Here's an example from Chekaluk et al.:

• Once a consensus set of copy number calls has been generated (Proposed Analysis: Copy number consensus calls #128; output should still be a SEG file), we will want to examine results by histology as specified by the title of this issue and Addition of R script to compare CNV caller output #142 (comment). Medulloblastoma is a good starting point because we have prior knowledge about what CNAs to expect (Figure 2 linked above by @jharenza).

• GISTIC files may be included in a later release (Addition of R script to compare CNV caller output #142 (comment)), but given the limitations of GISTIC, we would drop histologies with low numbers (Planned Analysis: Copy number plot showing recurrently amplified/deleted regions in different PBTA cancers #8 (comment)).

  If GISTIC files are included in a later release, we could use maftools, which is already on the Docker container, to produce plots: https://bioconductor.org/packages/release/bioc/vignettes/maftools/inst/doc/maftools.html#8_processing_copy-number_data

• We could also try using TAGCNA as an alternative to GISTIC. It is an R package on sourceforge rather than CRAN or Bioconductor that does not appear to be actively maintained and appears to be less frequently used compared to GISTIC.

---

Also I would note if anyone is going to continue development prior to #128, probably use CNVkit and not ControlFREEC (see caveat in README and the caveat in action in this plot).

[jaclyn-taroni]

GISTIC 2.0 is available as a GenePattern module. This requires a reference file:

Reference Genome File (-refgene) (REQUIRED)
The reference genome file contains information about the location of genes and cytobands on a given build of the genome. Reference genome files are created in MATLAB TM and are not viewable with a text editor. The GISTIC 2.0 release includes the following reference genomes: hg16.mat, hg17.mat, hg18.mat, and hg19.mat).

That description makes it seem like it might not be the most straightforward to obtain/derive?

Also requires:

Markers File (-mk) (REQUIRED)
The markers file identifies the marker names and positions of the markers in the original dataset (before segmentation). It is a three-column, tab-delimited file with an optional header. The column headers are:

Marker Name (marker name)
Chromosome (chromosome number)
Marker Position (in bases)

I think if we could run this in something like GenePattern Notebook so it's something public facing we can link to in a README, that could be good. Alternatively, something publicly facing on CAVATICA that can be linked to in the same way.

This requires a little bit more research before we know if it's feasible but wanted to report what I know so far.

[jharenza]

@jaclyn-taroni I may be able to run GISTIC tomorrow and provide in the data release, once I get the updated seg files.

[cansavvy]

I am working on getting this GISTIC plot set up in preparation for when the GISTIC data should be added to the release on v12.

[cansavvy]

Did we want this plot to be exactly like the example above? Here is what I have so far:

[cansavvy]

@jharenza , do you have thoughts about this rough draft GISTIC plot? What changes would you like to see?