Problem with inflated pseudocounts from EmpiricalBayesPrior

[Jayesh-Kumar-Sundaram]

EmpiricalBayesPrior=function(A,B) {

u=A>0 | B>0
A=A[u]
B=B[u]

x=median(log(A)-log(B))
y=max((quantile(log(A)-log(B),pnorm(1))-x)^2,(-quantile(log(A)-log(B),pnorm(-1))+x)^2)
if (is.infinite(x) || is.infinite(y)) {
    x=mean(log(A+1)-log(B+1))
    y=var(log(A+1)-log(B+1))
}
opt.fun=function(v) (digamma(v[1])-digamma(v[2])-x)^2+(trigamma(v[1])+trigamma(v[2])-y)^2
optim(c(1,1),opt.fun)$par

}

This is regarding the EmpiricalBayesPrior function. For some of my data it gives me a huge difference in pseudocounts as prior (For example: (0.16,208)). I see that I have good number of genes are detected only in B condition when compared to A condition in my case. I feel that it is probably biasing the estimation. Why did you use "x=median(log(A)-log(B))" first and then "x=mean(log(A+1)-log(B+1))" inside the if{}? I feel like having median function inside if{} might help my case. Any comments?