For the top hits from an R program, I was getting all zero p-values, which are not that useful for ranking or plotting.
Turns out that the R program was calculating the p-value in this manner:
1 - pchisq(71.12830,1)
But, according to the discussion here, in R
".Machine$double.eps is the smallest number such that 1+x can be distinguished from 1"
On my machine, we have that:
> .Machine$double.eps
[1] 2.220446e-16
So that is why this truncates to zero:
> 1 - pchisq(71.12830,1)
[1] 0
However, we can get a non-zero p-value if instead we compute it this way:
> pchisq(71.12830,1,lower.tail=FALSE)
[1] 3.347341e-17
