Issue with Substrate Prediction Consistency in PARAS - Same motif, different substrate prediction

[zainab629]

Dear PARAS Team,
I hope you are doing well.
I would like to report a potential issue I encountered while using your substrate prediction tool for NRPS A-domains.
The software appears to assign two different substrate predictions to the same A-domain motif. This inconsistency is unexpected, as a single A-domain signature sequence should ideally correspond to a single predicted substrate (or at least a clearly ranked prediction).
Could you please help clarify:
Whether this behavior is expected under certain conditions (e.g., ambiguity in the specificity-conferring code), or
If this might indicate an issue with the prediction algorithm or output formatting?
If needed, I would be happy to provide additional details about the sequence or analysis.
Thank you for your time and for maintaining such a valuable tool for the community.
Best regards,
Zainab

[davidmeijer]

Hi Zainab, thank you for trying PARAS! The signatures are indeed the same. Are the extended signatures the same as well? You can copy them by clicking the "ext. signature" buttons in the bottom right of each A-domain prediction and pasting them into another document to compare them.

[zainab629]

Thank you for your response and for pointing me to the extended signatures.

I compared the extended motifs, and they differ at two positions:
EGYSFDFSVDQTKVFGGGEHNSYGPAEISIASHK — predicted Alanine
QGYSFDFSVDQTKVFAGGEHNSYGPAEISIASHK — predicted Aspartic acid

While the core specificity-conferring residues are identical, the extended signatures differ at two positions (E→Q at position 1 and G→A at position 16).
I wanted to ask whether differences at just two positions in the extended motif are typically sufficient for PARAS to assign different substrates, or if this reflects a borderline/low-confidence prediction case. Additionally, are there confidence scores or thresholds associated with these predictions that could help interpret such discrepancies?
Thank you again for your help—I appreciate the clarification.

Best regards,
Zainab

[BTheDragonMaster]

Dear Zainab,

Thank you for using PARAS!

As PARAS is trained on the 34-residue code, it is absolutely possible that a different residue at a single position is sufficient to assign different substrates. However, it is very uncommon, and could reflect an error in the underlying dataset (possibly it is a datapoint we missed in our curation efforts; possibly there was an error in the publication the datapoint came from). It can also happen sometimes for A domains from fungi, for which the active site signature is usually not as predictive. From a quick database search I believe the domains you ran are indeed of fungal origin (see below).

I had a look at the 'query database' tab on the webpage (you can navigate to this yourself: select 'Substrate specificities by A-domain signature and input your signature), and found three training datapoints that share the signature you shared above: two recognise aspartic acid (the 1st A domain in BAZ95823.1 and EDU47082.1) and one alanine (the 1st A domain in ARP51711.1). I am confident this is why the model distinguishes between the two. I went into the literature (10.1021/acs.jnatprod.7b00183) to check if the annotation for ARP51711.1, which participates in phyllostictine biosynthesis, is correct. The authors confirmed through feeding experiments that the origin of one of the carbons of the amino acid is radio-labelled alanine, and therefore incorporated the recognition of alanine in their biosynthetic model. However, it may be possible that this alanine carbon is used in the biosynthesis of aspartic acid, which would mean that aspartic acid and not alanine is recognised by this domain. Possibly, the pathway would look something like this: alanine -> pyruvate -> oxaloacetate -> aspartate, which conserves the carbon skeleton, and hence the carbon the authors indicate (the carboxylic carbon of the backbone) could be originally derived from alanine but end up in aspartic acid, which may then be recognised by the A domain.

You are probably more familiar with the chemistry of your system, so if you think the A domain should recognise aspartic acid instead, please submit a correction to the database - PARAS is very much a community resource that we hope to keep improving over time!