Include values of zero in TSS metaplot?

Question

[goodez]

I am making a custom python script that finds average coverage at each position surrounding TSS. The script takes in a bedgraph file and a gtf file.

My question is: do you think ignoring regions of zero coverage will adversely affect my TSS plot? For example:

distance from TSS | coverage
-------------------------|--------------
-3000 | 0
-3000 | 2.3
-3000 | 0
-2999 | 0
-2999 | 0
-2888 | 3.1
-2888 | 2.9
-2888 | 2.1
-2888 | 0

It may seem like a strange question, but I need to change my workflow in order to consider the values of zero coverage. So would the overall trend of the metaplot be different if I ignore the regions of zero?

Thanks

Answer 1

[Friederike]

the extent to which this will matter will depend on the details of how you calculate the average coverage (which you haven't supplied). I also don't immediately get what the numbers shown above are supposed to mean - is every row a different sample? have you tried it out, i.e. have you generated plots with and without the zeros?

as a side note: there are a number of established tools to do these kinds of calculations, including NGSplot (R-based) and deepTools (python-based)

[goodez]

It is just the average for each position, so for position -3000 from TSS, my average would have been (0 + 2.3 + 0) / 3. Each row represents a different TSS (different gene). The coverage values are normalized per million. Sorry I realize it isn't totally clear but I did not want to go into the details of my python script.

I have been using an R package "ChiPseeker" which will give me a TSS plot but it does too much under the hood. I am doing my own script to have more control over the data. I'm going to redesign my workflow so I can use the zeroes.

Sorry for the weird question, I was just hoping to avoid including the zeroes.

[Friederike]

I was just hoping to avoid including the zeroes

If you already decided to exclude zeroes, what is your question about?

I am doing my own script to have more control over the data

deepTools' computeMatrix lets you tune numerous parameters, including the handling of zeros (--skipZeros) and the type of calculation (mean, median, ..., --averageTypeBins). plus it's fairly fast and optimized, extensively tested and widely used.

[goodez]

It's not that I decided to exclude them. My question was about the effect of including or excluding zeroes. I have the zeroes now and the reason they were missing before was because I was using a tool which was converting my bam files into coverage files in a way that skipped regions of zero coverage.

I'll admit I was just being a bit lazy with not wanting to go back and change my workflow. Thanks for telling me about computeMatrix! I have used deepTools but not tried the computeMatrix function.