Compare SignalP and Phobius (try to get better coverage of transit peptides and signal peptides)

[ValWood]

Once
#1195

done, compare
signalP (
Phobius (could use the InterPRoAPI for this)
DeepSig?
and decide thresholds to use to get best coverage.

We still have lots of signal peptides and transit peptides that are not found

Any other tools that could be included?

[ValWood]

Wen I run jackhmmer I get a signal peptide from phobius for https://www.pombase.org/gene/SPBC1E8.03c
(I think this is mainly were I am seeing them).
We don't have this from phobius in Pombase. So we need to find out what cut off InterPro uses.....

[ValWood]

SignalP:

One of the most widely used tools for signal peptide prediction.
SignalP 5.0, the latest version, utilizes deep learning techniques to improve accuracy.
It can differentiate between three types of signal peptides: Sec/SPI (bacterial), Sec/SPII (lipoprotein), and Tat/SPI (twin-arginine translocation).
Phobius:

Integrates signal peptide prediction with topology prediction.
Useful for distinguishing between signal peptides and transmembrane regions.
DeepSig:

Uses deep learning methods to predict signal peptides.
Claims to offer improved performance over traditional methods like SignalP.

Maybe we should also try DeepSig?

[ValWood]

So
i) what do we use for signal peptides at the moment?
ii) Is this page. displaying correctly ? https://www.pombase.org/term/SO:0000418 (I can access evidence codes/source)
iii) I thought we had more than 366 signal peptides?

[kimrutherford]

i) what do we use for signal peptides at the moment?

We get 214-ish from UniProt (via pombe-embl/external_data/uniprot_data_from_api.tsv) and the rest are loaded from pombe-embl/supporting_files/manual_so_term_annotations.tsv. Most of the manual annotations used to be in the contig files.

ii) Is this page. displaying correctly ?

Sorry! I've fixed the configuration and I'm re-releasing pombase.org now. The annotations will be back soon.

iii) I thought we had more than 366 signal peptides?

Those are the ones I know of.

[ValWood]

OK, I remember now we had a long discussion about the coverage being low

[kimrutherford]

OK, I remember now we had a long discussion about the coverage being low

There's a bit of discussion here:

• Create high confidence datasets for ER signal sequence (true positives and true negatives) #1195