The problem is there is no good definition of which TSS to use from a specific gene - I'm guessing for release 74 Ensemble simply reported the most upstream - but our research shows that is not a good proxy for the most used (only in ~50% of cases).

I think you might be better of with using the FANTOM5 TSS which are derived from high-thoughput experiments (CAGE) - you can read more about it here. In column2 (short description) the data derived TSS I linked to the TSS are annotated as pX@geneName. Here p1@geneName will correspond to the TSS most highly expressed across all F5 datasets - and might therefore be what you want.

Hope this helps.

Cheers Kristoffer

Also there is a published reference annotation data set (refTSS) here:

https://www.sciencedirect.com/science/article/pii/S0022283619302530 http://reftss.clst.riken.jp/

Ensembl Biomart does provide this here

Under attributes select chromosome name and transcript start

Then click "results" and download the TSV file