Casanovo v5.2.0 released

[bittremieux]

We are pleased to announce the release of Casanovo 5.2.0, which adds important new functionality for broader instrument support, more flexible training workflows, improved mzTab reporting, and more robust prediction behavior.

Highlights for users

Native timsTOF input support

Casanovo can now use Bruker timsTOF .d folders directly as spectra input files. This makes it easier to run Casanovo on timsTOF data without requiring a specialized intermediate conversion step.

No precursor m/z penalization

In de novo sequencing mode, Casanovo no longer filters or penalizes predictions based on precursor mass. This avoids penalizing otherwise plausible peptide predictions purely because of precursor mass disagreement. The precursor mass tolerance settings now apply only to database searching, where precursor mass filtering is still appropriate.

Improved mzTab output

The mzTab output has been updated to better follow the mzTab specification. In particular, the amino acid scores and ProForma peptide sequence columns have been renamed to their standardized mzTab-compatible names.

Casanovo also now tracks instrument-assigned scan numbers from MGF files and reports them in a dedicated mzTab column. This should make it easier to trace predictions back to the original spectra.

Users with downstream scripts that parse Casanovo mzTab files should check whether column names need to be updated.

Candidate peptide export during database searching

Database searching can now export a TSV file containing all candidate peptides using the --export flag. This is useful for inspecting search behavior, debugging, benchmarking, and downstream analyses.

Better support for fine-tuning and model development

Casanovo 5.2.0 adds improved support for training and fine-tuning workflows, including:

• loading all model states when resuming training;
• fine-tuning pretrained checkpoints with an extended residue vocabulary;
• per-file validation loss logging;
• tracking additional validation files to monitor catastrophic forgetting without affecting checkpoint selection.

These changes are especially useful for users adapting Casanovo to new instruments, datasets, or specialized training scenarios.

Updated software requirements

Casanovo now requires Python 3.10 or newer. Several core dependencies have also been updated, including Lightning and DepthCharge. Users upgrading from older environments may need to recreate or update their Python environment.

Bug fixes and usability improvements

This release also includes several fixes that improve reliability and reduce confusing warnings. Evaluation precision is now reported correctly, tokenizer warnings have been cleaned up, output overwrite checks are more complete, and short peptide predictions are handled more consistently.

[41ison]

Questions:

• is the 5.2.1 or future updates embedded in the CasanovoGUI?
• any plans of making the non-tryptic.ckpt compatible with new releases?

Thanks for developing this fantastic tool.

[wsnoble]

CasanovoGUI will automatically notice the latest version of the software next time you run it. At that point, you should be notified about the availability of a new version, and if you choose to, the software will automatically install the upgrade for you.

As for non-tryptic.ckpt, the following paper concluded that a single model can handle both tryptic and non-tryptic data:

https://pubs.acs.org/jprobs/article/23/10/4761/146152/Accounting-for-Digestion-Enzyme-Bias-in-Casanovo

Because of this, we no longer create separate models. The released weights are trained with multiple enyzmes.