Changelog

Changelog#

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased#

5.2.0 - 2026-06-02#

Added#

  • Support timsTOF files (as .d folders) as spectra input files.

  • Added --load_all_states flag to load all model states when resuming training.

  • A TSV file with all candidate peptides can be exported during database searching with the --export flag.

  • Track instrument-assigned scan numbers from MGF SCANS, SCAN, and SCAN ID header fields in a new opt_global_cv_MS:1003057_scan_number mzTab column.

  • Modified weights loading to match based on model selectors (e.g., orbitrap and timstof are currently supported) and major/minor versions.

  • Support fine-tuning a pretrained checkpoint with an extended residue vocabulary via the new_token_init config option.

  • Per-file validation loss logging via valid_CELoss/<stem> keys.

  • New --tracking_peak_path/-t CLI option for monitoring catastrophic forgetting on additional validation files without affecting checkpoint selection.

Changed#

  • Upgraded minimum Lightning version to 2.6.

  • Increased minimum Python version from 3.8 to 3.10.

  • Black version upgraded for Python 3.10.

  • Upgraded minimum DepthCharge version to 0.4.10.

  • Changed default gradient clipping to a norm of 1.0.

  • Updated train_batch_size documentation to reflect per-device/effective batch computation.

  • A more descriptive error message is logged for some annotated spectrum file parsing failure cases.

  • The precursor mass filter is no longer applied in de novo mode, and correspondingly peptide-level scores are no longer penalized based on the precursor mass. The config options precursor_mass_tol and isotope_error_range now only apply to database search mode.

  • The amino acid scores and ProForma columns in the output mzTab files have been renamed to opt_global_aa_scores and opt_global_cv_MS:1003169_proforma_peptidoform_sequence, according to the mzTab specification.

  • Minor speedup during database searching through optimized candidate selection.

Fixed#

  • A mismatching parameter warning will now only be triggered for the tokenizer if the config and checkpoint tokenizers do not have equivalent vocabularies.

  • Removed erroneous tokenizer vocabulary warning.

  • Fixed an issue which led the reported peptide precision to be 0 during evaluation mode.

  • Peptide predictions failing the minimum peptide length are not reported, irrespective of whether they match or exceed the precursor mass.

  • Setting --output_root to a directory will no longer cause an error.

  • The --force_overwrite flag now also checks whether mzTab output files would be overwritten.

  • Fixed an issue where some predictions that are one residue less than the configured minimum peptide length are reported.

5.1.2 - 2025-12-11#

Changed#

  • Hotfix to ensure compatibility with PyTorch v2.6 when loading weights files.

5.1.1 - 2025-11-27#

Changed#

  • Hotfix to avoid excessively slow peptide parsing during mzTab export.

5.1.0 - 2025-10-22#

Changed#

  • Avoid redundant spectrum encoding during database mode to speed up PSM scoring.

  • accelerator: "auto" is overwritten to accelerator: "cpu" on Apple Silicon devices due to MPS compatibility issues.

  • Removed reverse_peptides from the configuration since decoding from the C-terminus to the N-terminus yields better results.

  • Log a warning if residues in the user configuration are not present in the tokenizer alphabet.

  • The full ProForma sequences of predicted peptides are reported in the op_ms_run[1]_proforma of the output MzTab file for backwards compatibility.

Fixed#

  • Inline mass modifications are now specified separately from the sequence column in the modifications column of the output mzTab file, as per the mzTab specification.

5.0.0 - 2025-07-09#

Added#

  • Casanovo-DB mode (casanovo db_search) to use Casanovo as a learned score function for sequence database searching (given a FASTA protein database).

  • During training, model checkpoints will be saved at the end of each training epoch in addition to the checkpoints saved at the end of every validation run.

  • Besides as a local file, model weights can be specified from a URL. Upon initial download, the weights file is cached for future re-use.

  • Training and optimizer metrics can be logged to a CSV file by setting the log_metrics config file option to true. The CSV file will be written to under a sub-directory of the output directory named csv_logs.

  • New configuration options for detailed control of the gradients during training (gradient accumulation, clipping).

  • New configuration option min_peaks to discard low-quality spectra with too few peaks.

Changed#

  • Removed the evaluate sub-command, and all model evaluation functionality has been moved to the sequence command using the new --evaluate flag.

  • The --output option has been split into two options, --output_dir and --output_root.

  • The path suffix (extension) of --output_root will no longer be removed as it was with the old --output option.

  • The --validation_peak_path is now optional when training; if --validation_peak_path is not set then the train_peak_path will also be used for validation.

  • The tb_summarywriter config option is now a boolean config option, and if set to true the TensorBoard summary will be written to a sub-directory of the output directory named tensorboard.

  • Input peak files can now be specified as both individual file(s) and a directory.

  • Peptidoforms are specified using ProForma 2.0 notation by default.

  • DepthCharge is upgraded to the latest version 0.4.8.

  • The product of the raw amino acid scores is used as the peptide score, rather then the arithmetic mean.

  • Amino acid scores are directly reported, rather than averaged with the peptide score.

  • The amino acid-level score of stand-alone N-terminal modifications is combined with that of the leading N-terminal residue.

  • Renamed the n_peaks configuration option of the maximum number of peaks to retain in a spectrum to max_peaks.

  • Beam search decoding has been optimized for computational efficiency, achieving increased prediction speed.

Fixed#

  • Precursor charges are exported as integers instead of floats in the mzTab output file, in compliance with the mzTab specification.

  • Fixed log entries written to the config file instead of the log file when running the configure command.

Removed#

  • Removed the save_top_k option from the Casanovo config, the model with the lowest validation loss during training will now be saved to a fixed filename <output_root>.best.ckpt.

  • The model_save_folder_path config option has been removed; model checkpoints will now be saved to --output_dir during training.

4.3.0 - 2024-12-13#

Fixed#

  • Amino acid scores in the mzTab output were reported in reversed order.

4.2.1 - 2024-06-25#

Fixed#

  • Pin NumPy version to below v2.0 to ensure compatibility with current DepthCharge version.

4.2.0 - 2024-05-14#

Added#

  • A deprecation warning will be issued when deprecated config options are used in the config file or in the model weights file.

Changed#

  • Config option max_iters has been renamed to cosine_schedule_period_iters to better reflect that it controls the number of iterations for the cosine half period of the learning rate.

Fixed#

  • Fix beam search caching failure when multiple beams have an equal predicted peptide score by breaking ties randomly.

  • The mzTab output file now has proper line endings regardless of platform, fixing the extra \r found when run on Windows.

4.1.0 - 2024-02-16#

Changed#

  • Instead of having to specify train_from_scratch in the config file, training will proceed from an existing model weights file if this is given as an argument to casanovo train.

Fixed#

  • Fixed beam search decoding error due to non-deterministic selection of beams with equal scores.

4.0.1 - 2023-12-25#

Fixed#

  • Fix automatic PyPI upload.

4.0.0 - 2023-12-22#

Added#

  • Checkpoints include model parameters, allowing for mismatches with the provided configuration file.

  • accelerator parameter controls the accelerator (CPU, GPU, etc) that is used.

  • devices parameter controls the number of accelerators used.

  • val_check_interval parameter controls the frequency of both validation epochs and model checkpointing during training.

  • train_label_smoothing parameter controls the amount of label smoothing applied when calculating the training loss.

Changed#

  • The CLI has been overhauled to use subcommands.

  • Upgraded to Lightning >=2.0.

  • Checkpointing is configured to save the top-k models instead of all.

  • Log steps rather than epochs as units of progress during training.

  • Validation performance metrics are logged (and added to tensorboard) at the validation epoch, and training loss is logged at the end of training epoch, i.e. training and validation metrics are logged asynchronously.

  • Irrelevant warning messages on the console output and in the log file are no longer shown.

  • Nicely format logged warnings.

  • every_n_train_steps has been renamed to val_check_interval in accordance to the corresponding Pytorch Lightning parameter.

  • Training batches are randomly shuffled.

  • Upgraded to Torch >=2.1.

Removed#

  • Remove config option for a custom Pytorch Lightning logger.

  • Remove superfluous custom_encoder config option.

Fixed#

  • Casanovo runs on CPU and can pass all tests.

  • Correctly refer to input peak files by their full file path.

  • Specifying custom residues to retrain Casanovo is now possible.

  • Upgrade to depthcharge v0.2.3 to fix sinusoidal encoding and for the PeptideTransformerDecoder hotfix.

  • Correctly report amino acid precision and recall during validation.

3.5.0 - 2023-08-16#

Fixed#

  • Don’t try to assign non-existing output writer during eval mode.

  • Specifying custom residues to retrain Casanovo is now possible.

3.4.0 - 2023-06-19#

Added#

  • every_n_train_steps parameter now controls the frequency of both validation epochs and model checkpointing during training.

Changed#

  • We now log steps rather than epochs as units of progress during training.

  • Validation performance metrics are logged (and added to tensorboard) at the validation epoch, and training loss is logged at the end of training epoch, i.e. training and validation metrics are logged asynchronously.

Fixed#

  • Correctly refer to input peak files by their full file path.

3.3.0 - 2023-04-04#

Added#

  • Included the min_peptide_len parameter in the configuration file to restrict predictions to peptide with a minimum length.

  • Export multiple PSMs per spectrum using the top_match parameter in the configuration file.

Changed#

  • Calculate the amino acid scores as the average of the amino acid scores and the peptide score.

  • Spectra from mzML and mzXML peak files are referred to by their scan numbers in the mzTab output instead of their indexes.

Fixed#

  • Verify that the final predicted amino acid is the stop token.

  • Spectra are correctly matched to their input peak file when analyzing multiple files simultaneously.

  • The score of the stop token is taken into account when calculating the predicted peptide score.

  • Peptides with incorrect N-terminal modifications (multiple or internal positions) are no longer predicted.

3.2.0 - 2022-11-18#

Changed#

  • Update PyTorch Lightning global seed setting.

  • Use beam search decoding rather than greedy decoding to predict the peptides.

Fixed#

  • Don’t use model weights with incorrect major version number.

3.1.0 - 2022-11-03#

Added#

  • Matching model weights are automatically downloaded from GitHub.

  • Automatically calculate testing code coverage.

Changed#

  • Maximum supported Python version updated to 3.10.

  • No need to explicitly specify a config file, the default config will be used instead.

  • Initialize Tensorboard during training by passing its directory location.

Fixed#

  • Don’t use worker threads on Windows and MacOS.

  • Fix for running when no GPU is available.

3.0.0 - 2022-10-10#

Added#

  • The first PyPI release! :tada:

  • Tests are run on every PR automatically.

  • Test code coverage must be maintained or improved with each change.

  • Log the active Casanovo configuration.

  • Log to both the console and a log file.

  • Use all available hardware resources (GPU and CPU).

  • Add ICML paper citation info.

  • Document GPU out of memory error in the README.

  • Allow mzML and mzXML peak files as input during predicting.

  • Ability to reuse an existing HDF5 index during training.

  • Move the changelog information from the README to CHANGELOG.

Changed#

  • Consistent code formatting using black.

  • Assign a negative score to peptide predictions that don’t fit the precursor m/z tolerance.

  • Faster empty token detection during decoding.

  • Consistently set the random seed to get reproducible results.

  • Spectrum indexes are written to temporary HDF5 files.

  • Use spectrum_utils for spectrum preprocessing.

  • Rename the mode to predict peptides for unknown spectra from test to predict.

  • Export spectrum predictions to mzTab files.

  • Update the residue alphabet to include N-terminal modifications (acetylation, carbamidomethylation, NH3 loss).

  • Specify input peak files as a shell pattern rather than by their directory.

  • Make the config file optional to specify.

Removed#

  • Always preprocess spectra, rather than having this as a user option.

Fixed#

  • Don’t log overly detailed messages from dependencies.

  • Don’t crash on invalid spectrum preprocessing.

  • Ensure that config values have the correct type.

  • Don’t crash when an invalid residue is encountered during predicting (i.e. an N-terminal modification in another position).

  • Don’t penalize peptide scores if the precursor m/z fits a C13 difference.

2.1.1 - 2022-07-27#

Added#

  • Update tutorial in the README.

2.1.0 - 2022-07-02#

Fixed#

  • Use latest depthcharge version with stable memory usage and fix to positional encoding for amino acids.

2.0.1 - 2022-06-13#

Added#

  • Include release notes in the README.

2.0.0 - 2022-06-05#

Added#

  • Additional CLI functionality.

  • Unit testing using pytest.

  • Include a tutorial in the README.

  • Publish documentation using sphinx/ReadTheDocs.

Changed#

  • Specify config as a YAML file.

1.2.0 - 2022-03-07#

Added#

  • Include peptide and amino acid confidence scores in output file.

1.1.2 - 2022-02-20#

Added#

  • Support for multiple input MGF files in a directory.

1.1.1 - 2022-02-10#

Added#

  • Provide more CLI options.

  • Ability to specify a custom config file.

1.1.0 - 2022-02-04#

Added#

  • Data infrastructure.

  • Model and training/testing functionality.

1.0.0 - 2022-01-28#

Added#

  • Initial Casanovo version.

Contents