Best practice for non‑tryptic de novo sequencing (ZenoTOF, no fixed modifications)

[aclguerreiro]

Hello everyone,

I would like some guidance on the recommended way to configure Casanovo for my use case, as I am seeing unexpected behavior when deviating from the default settings.
I became aware of Casanovo through a colleague who attended a conference in Antwerp, where it was presented and discussed.

Setup

Casanovo v4.2.0
Model: casanovo_nontryptic.ckpt
Non‑tryptic peptides, de novo sequencing
Data acquired on a SCIEX ZenoTOF
Converted to MGF (mzML caused issues)

What works

Running Casanovo with the default configuration works well: the pipeline completes successfully and the resulting confidence scores for the peptide of interest are generally reasonable (≥0).

What I tried to change

• Reduced precursor mass tolerance from the default (~50 ppm) to 10 ppm, which seems more appropriate for ZenoTOF data.

• Attempted to remove fixed modifications in particular C+57.021, since no alkylation or similar sample preparation was performed.This was done by modifying the "amino‑acid/modification vocabulary" and restricting modifications to the "amino acid modifications" section.

Issue

After these changes, Casanovo still runs, but the scores for the peptide of interest are all below 0.

Questions

• Am I using the correct setup and configuration strategy for this goal?

• Is it valid to remove fixed modifications in the way I attempted?

• For a truly modification‑free setup, is it recommended to keep the default configuration and interpret results accordingly?

• More generally, is reducing tolerance_ppm to 10 ppm advisable for ZenoTOF data, or is the default tolerance intentionally loose?

Since I do not have a strong bioinformatics background, I would greatly appreciate any guidance.

Thanks in advance, and thank you for developing Casanovo.

Best regards,
Ana

[wsnoble]

Hi Ana,

Thanks for your interest in Casanovo. I'm glad to hear that it is at least working in the default settings. The short answer to your question is that you can't change Casanovo's alphabet (e.g., removing the fixed modification on C) without retraining the model from scratch. There is a recent paper that demonstrates how to allow this kind of flexibility in a Casanovo-like model (Modanovo):

https://www.sciencedirect.com/science/article/pii/S1535947625006000

We intend to incorporate functionality like this into Casanovo in the near future. But even in that case, you would need to do some fine-tuning of the model using your adjusted alphabet.

As an aside, we have found that the precursor filter is not really necessary, so this will be removed in the next release of Casanovo. You are welcome to change from 50 ppm to 10 ppm, but the only change this will effect is to subtract 1 from the scores of spectra whose predictions have ppm differences in the range 10-50. Empirically, we observe that Casanovo does a good job of assigning low scores to out-of-range predictions, making this filter unnecessary.

Cheers.
Bill

[aclguerreiro]

Hi Bill,

Thank you very much for the clear explanation, it was very helpful.
I also appreciate the pointer to Modanovo and the note that similar flexibility may be added to Casanovo in the future.

For now, I will proceed with the default configuration and vocabulary and interpret the results accordingly.

Thanks again for your time and for developing Casanovo.

Cheers,
Ana