search

saezlab / liana-py

LIANA+: an all-in-one framework for cell-cell communication
http://liana-py.readthedocs.io/
GNU General Public License v3.0
169 stars 21 forks source link

Should we exclude CellTalkDB from curated resources list? #75

Open dbdimitrov opened 1 year ago

dbdimitrov commented 1 year ago

@deeenes I keep finding strange interactions that occur as PPIs but not actual CCC and are reported as such in CellTalkDB - e.g. SORBS1-INSR or interactions with SMAD3. Might be best if we exclude CellTalkDB from the curated resources list in OmniPath, and hence LIANA.

dbdimitrov commented 1 year ago

PS. It seems to me that the authors of the resource, took some resources (specifically LRdb as they have a huge overlap and this one is not curated) at face value and then curated them i.e. found the literature support for the interaction, but did not necessarily check if it occurs in the context of CCC.

dbdimitrov commented 1 year ago

At the very least we should exclude the interactions the reference of which points to LRdb/SingleCellSignalR and Ramilowski as these are not curated, simply transferred.

dnjst commented 11 months ago

I can give some feedback as a user on the interaction lists.

The old liana consensus was coming up with a lot of spurious interactions like cofactors as receptors. Manually cutting down to only interactions from CellPhoneDB v4 | CellChat | Connectome got rid of those.

Then, with release of the new CellPhoneDB v5 (manually curated), it seems to be high quality and incorporates almost all of the "good" interactions from other manual resources. It also has useful metadata. Since it is under active development (unlike others which publish and then static), I have had good results with using that as the baseline, and don't bother trying to get more interactions (or if I find more in the literature, submit to CPDB).

dbdimitrov commented 11 months ago

Thanks @dnjst, this is helpful to hear.

We will soon make the latest version of CPDB5 available via liana (and will be updated on each version). So, that it's also versioned and easily accessible. Though, I would also consider setting CPDBv5 as default (under the assumption that while it might be missing interactions which might be of interest, those in it reliable). Or at the very least, I will rely more on CPDB when I rebuild the consensus.

Daniel