Problems in dividing genome to non-overlapping bins and count genes

Question

[milk841103]

Hello,

I was working on analyzing the distribution of CNV-affected genes in a genome, so I divided the genome into non-overlapping bins with same length and looking at the number of CNV-affected genes within each bin. the method I used was from this post: C: Finding gene density from reference genome using R As you can see, the GRange table containing location of all the bins had an extra column added showing number of genes (in my case, CNV-affected genes) in each bin. I then found the 10 bins that contains the most CNV-affected genes, extracted their coordinates and tried to ID the individual genes (from a list of CNV-affected genes that I generated) that locate in those bins for further analysis. the code I used to extract these genes was written in unix command from this post I posted: http://unix.stackexchange.com/questions/303809/select-multi-column-rows-based-on-ranges-specified-in-a-separate-file

however, I found that while there are 247 CNV-affected genes identified in these 10 bins using the first method in R, my unix command suggested otherwise as only 175 CNV-affected genes were found in the same 10 bins. my co-worker wrote a perl script for the same purpose and the results also showed 175 genes were found.

while using the R method, I found that in some bins the number of genes called by function countOverlaps was off by 1 when varifying using the subset function (both are mentioned in the first post) but I assume it was because genes crossing two bins, however i don't understand why the results I got from R and unix command could differ so much. Can anyone help explaining this issue? Thank you very much!!

Answer 1

[Alex Reynolds]

BED or tab-delimited text files you work with via Unix likely use a half-open 0-based index. Grange objects use a closed 1-based index. So this could potentially create one-off errors if you're not adjusting coordinates before trying to do an apples-to-apples set operation.

[milk841103]

thank you for your answer! do you mind telling me how should I adjust coordinates? and also could this 1-off error make such huge defference? because the sums of genes from 10 bins calculated using R and unix/perl script were off by 50...

[Alex Reynolds]

It will depend on how you count things, but generally, you might start with the following overview and decide if this is an issue. You could start by testing overlaps on a handful of ranges via R/grange and Unix/Perl, to see if you get the same or different answers with whatever procedures you're using. Cheat Sheet For One-Based Vs Zero-Based Coordinate Systems