[rebrendi]

Unless you're downloading hundreds of genomes, it's not a problem to quickly check a couple genes to make sure the dataset is what you think it is. That's a good thing to do anyway for any dataset you don't produce yourself. Frankly, you could have done that between when you wrote your last message and my reply.

I am not that good in this particular topic, and it is not a question of minutes anyway. It took me several weeks just to realize (by chance!) that something was going wrong with Yeast.

Now, with respect to other datasets, your advise to check a random gene from the dataset is based on the assumption that all the entries of the same dataset behave the same. This assumption is based on my example with S.Cerevisae genome above. However it could be (and I still suspect that this could be the case) that some entries of the mouse or human datasets could indicate the 5'-UTR as "gene start", whereas other entries for which the 5'-UTR information is simply not available, indicate the ORF start as the "gene start". Thus checking a random gene wound not prove anything. I hope it is not the case, but just trying to make it sure.

[rebrendi]

I have found another TSS database: http://dbtss.hgc.jp/index.html
Again, the coordinates do not coincide for several manually checked genes.
Is it really that difficult to determine TSS experimentally?
What's the best TSS database, anyway?