Question About Extracting Species-Specific Peptides from Unipept Database

[baynec2]

Hi,

First of all, thank you for all the work you've done on Unipept — it's an incredible tool, and it’s been super helpful for our metaproteomics work!

I’m currently trying to generate a list of peptides that are unique at the species level by querying the Unipept database. I expected this information might be available in the sequences.tsv.gz file from the 2025-03-19 release, since my understanding was that this file contains peptides and their corresponding lowest common ancestor (LCA).

However, when testing peptides from that file against the web interface, I noticed mismatches. For example, the peptide AAGGQGLHVTAL has an LCA of 9606 (Homo sapiens) in the file, but the web tool returns an LCA of Mammalia, with 64 UniProt matches. This seems to indicate that the LCA in sequences.tsv.gz is not calculated the same way as in the web version — or perhaps I’m misunderstanding the structure of the file or working with outdated information.

Could you help clarify where I might find accurate species-level unique peptides or how to interpret the LCA field in the file correctly?

Thanks so much for your time!

Best,
Charlie

[pverscha]

Hi Charlie,

Thank you for contacting us. The mismatches you're seeing are the result of two important differences between the sequences file that you've downloaded and the Unipept web interface running on our website:

• The generated sequences.tsv.lz4 is created from the SwissProt-entries in UniProtKB only, where our web interface uses all proteins from TrEMBL + SwissProt.
• These sequences are tryptic, where our current web interface also matches with non-tryptic peptides in the original proteins. Even though the input peptides can be tryptic, there will still be more matches if you match them in a non-tryptic way.

To clarify this last point, here is a small example:

Input peptide:
AALTER

Database:
Prot1 -> XXXXRAALTERYYYY
Prot2 -> XXRXXAALTERYYY

With the old tryptic matching system, only Prot1 will be matched by Unipept because all proteins are in-silico digested beforehand.

The new system, however, will also match with Prot2 because AALTER is present in Prot2, even though it is not preceded by an R or a K.

Since we've had more people asking about the unique peptides for a taxon, I've developed a small script that emulates the behavior of our old "unique peptide finder" using our API: https://github.com/unipept/unipept/wiki/Emulating-the-Unique-Peptide-Finder

I'm not 100% sure if the script still works. Please let me know if it doesn't, I will update it then.

Kind regards,

Pieter

[baynec2]

Hi Pieter,

Thank you for the clear explanation—that makes a lot of sense.

As a follow-up, is there a way to obtain a file similar to sequences.tsv.lz4, but containing peptides derived from both Swiss-Prot and TrEMBL entries?

For context, I would like to identify unique tryptic peptides across all species, rather than focusing on a specific taxon.

Thanks again for your help.

Best regards,
Charlie

[pverscha]

Hi Charlie,

We used to generate this file for every new release of UniProt. However, since the Unipept metaproteomics pipeline no longer depends on predigested proteins, this is no longer the case. The file is still generated, though, as a side-effect if you build the indices for our metagenomics pipeline.

You can download this repository (unipept-databasE), and navigate to the folder scripts. There, you'll find a script generate_umgap_tables.sh that will produce the sequences.tsv.lz4 file for TrEMBL (or SwissProt, or both, up to you).

Documentation on how to use this script can be found here: https://github.com/unipept/unipept-database/wiki/Building-the-UMGAP-indexes.

You will need to run it in tryptic mode, and as a side-effect of running this script, the sequences file will be generated.

Note, this script does require a substantial amount of storage space (~ 1TiB) to run, and a list of dependencies:

• rust
• curl
• uuidgen
• lz4
• pigz
• pv
• umgap (can be installed through https://unipept.ugent.be/umgap)

[baynec2]

Hey Pieter,

That's great! Thank you for the help. I will give it a try as soon as I get the chance.

Best,
Charlie