Coverage depth vs mutation frequency

[lbeltrame]

Hello,

the group I am working in is preparing for some targeted (~50 genes) analysis of variations on tumor samples using the Ion Torrent platform. As the number of chips needs to be minimized due to cost, we were trying to design the experiment to obtain the best results out of our limitations.

This brings us to the problem: determining the optimal coverage depth per mutation frequency. In other words, what is a good coverage depth to detect mutations with these frequencies:

• 20%
• 10%
• 5%

Is there any publication or reference where I can get an ideal grasp of this so we can adjust the experiment design accordingly? Thanks in advance.

[rama]

hi, just wondering if you ever got through figuring this out. I would really appreciate your thoughts on how we could address the same problem that we have right now.

thanks

[SNPsaurus]

In my simple view of statistics, I would calculate it like this:
A mutation at 10% frequency has a .1 chance of being seen by a read.
So a .9 chance of not being seen. The chance of it not being seen in 10 reads is .9^10 = .35 (35%).

To see 95% of the mutations at 10% frequency, you would solve:
.05 (5% chance of not being seen) = .9^X
X = log .05 / log .9 = 28X coverage.

This is idealized and doesn't take into account variation in coverage, etc. Unfortunately, if the average coverage is 28X, then regions of higher coverage won't have much higher chances of seeing the mutation (40X coverage will have a 99% chance instead of a 95% chance), but regions of lower coverage will have much worse performance (15X coverage will see only 80% of the mutations).