For a recent analysis using glmTreat, I get a couple of unadjusted p-values > 1. This doesn't particularly concern me, as these genes are obviously not of interest, but given that these are not valid I thought I would ask if this is expected.

Stepping through glmTreat and looking at z.tau and z.tau2, I get this, in case it helps:

Browse[2]> ind <- which(p.value > 1)
Browse[2]> z.tau2[ind]
[1] 1.183449 1.653373
Browse[2]> z.tau[ind]
[1] 1.186449 1.653595
Browse[2]> 2 * fun(-z.tau2[ind], (z.tau*sgn)[ind])
[1] 1.000967 1.000061

Thanks for reporting this. You're right. These p-values are not expected.

The glmTreat is designed for two-sided hypothesis tests. The reason for such unexpected behaviors is that the observed fold change (FC) is closer to one FC boundary (threshold) but its likelihood is closer to the likelihood evaluated at the opposite FC boundary. This would be possible when the observed logFC is very close to 0 and the FC threshold value is large. It wouldn't affect results for genes with large FCs.

Nevertheless it is still a bug, and I'll fix it accordingly.