Normalizing values for computing log ratios

[aprice67]

I am currently working on a project with RNA-Seq data. I need to compute a ratio of the number of reads at each position for two different protocols so I have log(protocol1value/protocol2value). The problem is that its quite often I will have something like log(25/0), which is impossible to compute. Is there a way to normalize the values so that I can make this sort of computation accurately?

I want to add that, based on advice by some colleagues, I have considered and dismissed the possibilities of adding 1 to all counts before computation or replacing all 0's with very small numbers like .000001.

I'm sure there must be a way, I am just having trouble finding it. Thanks in advance.

[jparsons]

There is no accurate computation for log(25/0). If there was one, your computer would be happy giving it to you.

Your colleagues have provided you with the appropriate workarounds for this problem. They aren't ideal, but they're all you can do. It's just a matter of asking yourself how to represent your 25/0.

You have no real way of knowing what the "true" ratio of counts for that particular position is, so you need to choose whether you should underestimate the ratio (25/1), probably overestimate the ratio (25/1e-6), or put a big shiny asterisk next to them all and say "I don't know the real answer and am not comfortable guessing"