delta_mass_ppm missing in local installation

[07liesin]

Hello,

I recognized that there is no delta_mass_ppm in the output csv of the locally installed InstaNovo compared to the online huggingface version. It is also missing in the documentations first prediction showcase. But later on it is recommended to filter the results according to the log_probabilities and delta_mass_ppm columns.

How can I get the delta_mass_ppm in my results with the local installation?

Thank you,
LL

[rcatzel]

Hi @07liesin,

In our current codebase the column delta_mass_ppm should be written to the output csv according to this code. Could you please check that your local version of InstaNovo is up to date with the main branch?

[07liesin]

Hi Rachel, I checked it in my 1.1.2 installation and also updated it to 1.1.4. The respective lines are there but not in the diffusion predict.py. If I run transformer model only (with instanovo transformer predict...) I now get the delta_mass but not if I want to include the refinement via the diffusion model (with instanovo predict...). But on huggingface it seems to work on both models,although the delta_mass results in the value ~5000,which looks strange to me(while using the sample data mgf from github and default settings on huggingface). Thank you for your answer!

…

________________________________ Von: Rachel Catzel ***@***.***> Gesendet: Freitag, 13. Juni 2025 11:11:31 An: instadeepai/InstaNovo Cc: Liesinger, Laura; Mention Betreff: Re: [instadeepai/InstaNovo] delta_mass_ppm missing in local installation (Issue #108) [https://avatars.githubusercontent.com/u/55193541?s=20&v=4]rcatzel left a comment (instadeepai/InstaNovo#108)<#108 (comment)> Hi @07liesin<https://github.com/07liesin>, In our current codebase the column delta_mass_ppm should be written to the output csv according to this<https://github.com/instadeepai/InstaNovo/blob/main/instanovo/transformer/predict.py#L328> code. Could you please check that your local version of InstaNovo is up to date with the main branch? — Reply to this email directly, view it on GitHub<#108 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BM5CPYZNJF6K3QUORRRVHPL3DKIUHAVCNFSM6AAAAAB6DFP5GGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDSNRZGY2TMOBZG4>. You are receiving this because you were mentioned.Message ID: ***@***.***>!

[07liesin]

Hi again,

I now calculated the delta masses of both output peptides (diffusion and transformer, including given modification) on my own. When I compare my transformer delta masses with the ones from instanovo there is just an overlap of 15% of all output sequences.

Could be that I miscalculated due to missing information of your calculation.
But that leaves me at the moment with the choice between taking only transformer model for evaluation (where I can use instanovos delta-mass-ppm) or just 15% of the best peptides according to the probabilities before even filtering on delta-mass or probabilities.

So any information in regard of your delta-mass calculations or simply an implementation of fully calculated delta-masses in the next versions would be very helpful.

Here some additional remarks: I also found peptides now (a few hundred), which have the final log_prob from the diffusion model but no assigned peptide sequnence, which makes them unusable:

Could you please also explain what the new column "precursor_mass_match" means in detail?

Thank you for your answer,
LL