required coverage calculation and statistical analysis

[alezhe]

hi
We are studying genetic variation in populations. So we mix genomic DNAs of several individuals and use it as template to amplify regions of interest.

According to Roche manual

40x coverage will result in greater than 99.9% chance of detecting a heterozygote

What to do if variation frequency is not 0.5 (heterozygote), but, let’s say 0,01 (one heterozygote in 50 individuals).

According to the same manual

1% variation of single base changes or multibase deletions 5000x coverage will average 50 reads for a typical variant

It’s obvious that in case 5000 coverage we expect 50 reads with polymorphic nucleotide.
In case coverage 500 we must expect 5 polymorphic reads, but have a good chance to get 2-3 polymorphic reads only (and I don’t take into account mixing error), so it might be interpreted as sequencing error as well.

I’m looking for statistical approaches that will allow define desired coverage and interpret real data to be sure (with some good chance, of course) whether a distinct alignment mismatch is a real polymorphism or just errors.

Does somebody know usefull links?
Thanks.

[Heisman]

There are pooled sequencing papers out there (SNPSeeker/SPLINTER, CRISP, etc); I'd check those and see what they say.