Implement G correction for CAGEexp class

[charles-plessy]

Hi Damir, I am exploring the idea of keeping the count of how many extra Gs were removed at each CTSS. Do you by chance have some good test data that the whole CAGEr package could use for examples and regression tests in all the functions related to G correction ? (And maybe we could even replace the example in the vignette with one using that data…)

[sga91]

What would be a good test data? I am currently analyzing the John Lys procap data with CAGEr, they quality is really good.

[charles-plessy]

Thanks for offering your help! Ideal test data should be:

• from an organism with a small genome, for which a BSgenome package exists. The reason is that users and autobuilders will need to install this package to run the tests. The current dataset, aligned on an old version of the Zebrafish genome, is not ideal for that.
• a subset of a good published dataset in which CAGEr has been used. The raw data should be reduced as much as possible to save space, because Bioconductor source packages should not be larger than 5 MiB, and regression tests and vignettes should run fast. This leaves just enough space for a few promoter regions (we need to figure out what is the bare minimum for testing the SOM and promoter shape functions).
• located on both strands and at least two different chromosomes (the current test data is only on a single chromosome). The goal is to have more chance to catch bugs, especially with parallelised loops. Maybe having a few tag on the first/last base of some chromosomes (think chrM for instance) would also help catching bugs.
• comprise a 4~5 samples like the current test data, for testing plot functions, sample merging, etc. To catch bugs with zero values, promoters in some samples should have a null expression in some tag clusters and annotation ranges.
• in BAM format, with ~80% G-addition, so that we can test the extraG functions (current test data is in ctss format).

Bonus points if the dataset has been clearly dedicated to the public domain (CC0 licence for instance) by its laboratory.

Please let me know if I missed something important in the wishing list :)

[sga91]

I feel the procap data from Drosophila S2 cells would be a good choice then, the only issue I see is the number of samples - there are two replicates but only in control conditions.

[charles-plessy]

@snikumbh, @da-bar, what do you think about this ?

[da-bar]

Hi Charles,

sorry I haven't replied sooner. I think keeping track of the number of removed Gs is a great idea. We are currently really looking into the G removal, and how it impacts the downstream analysis, etc.

Regarding the datasets, I can list a couple of them out of my mind, but I am positive if we don't manage to identify an adequate one, I will be able to dig out other.

Regarding the points you made:

We have the existing dataset (extended by some more developmental stages using nAnTi-CAGE) mapped to the newer version of zebrafish (danRer10 and danRer11). If we think we can continue using zebrafish, I can either subsample the bams, or use a single chromosome (which is not preferred as you stated). The benefit of continuing to use zebrafish is that the downstream part of the vignette intuitively works well with zebrafish embryonic development. I am not so familiar with Drosophila (probably @snikumbh is more familiar in those than me) or C. elegans samples, which have small genomes. Also, a good candidate that follows all the points from your list are certainly the yeast samples from the SLIC-CAGE paper.

Please let me know what you think, and we can proceed then from there.

Best, Damir

[sga91]

I forgot that the purpose of the analysis was to include it in the vignette and it that case I agree that it makes sense to continue with zebrafish, which I'm not familiar with. The yeast samples may be a good idea. There is also this recent paper from Margaret Fuller in which they perform CAGE-seq in the Drosophila male germline, two stages, from spermatogonia to spermatocytes and in at least two conditions. Also note that Drosophila genome is overall GC richer than yeast and zebrafish.