How much would you trust TMHMM, SignalP predictions?

[yealing]

Hi all,

In a recent mini-project, my supervisor and I talked through this matter. The mini project was set to functionally annotate a set of predicted genes in silico. We subjected the amino acid sequences through several Bioinformatics tests. Among them, TMHMM for transmembrane region prediction and SignalP for signal peptide and signal anchor prediction.

Upon completion of the tests, we discussed the boudary for declaring a predicted gene as functional gene/possible gene with unknown function. And the main question is: as seen above, how much can you trust TMHMM & SignalP results?

My point of view is that these predictions can only be used to support those of homology-based annotation. Genes without positive results in homology-based tests cannot be regarded as functional genes even if they came up positive for any of these tests.

My supervisor seemed to think that it is quite difficult for just any random sequence to match the criteria for either one of them (transmembrane/signal sequence), therefore genes with positive results can be regarded as functional genes.

Any opinion are welcomed.