Skip to content

v4.0.0

Choose a tag to compare

View all tags
@RobbinBouwmeester RobbinBouwmeester released this 24 Jul 09:18
· 63 commits to main since this release
v4.0.0
5dfb7e3
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

DeepLC v4 brings a new top-performing model trained on nearly 1000 different LC setups simultaneously. This ensures that the default model works well on almost any dataset, even with simple calibration, or with model fine-tuning if more reference data is available. Aside from the new model, the code base has been completely redesigned to be more user- and developer friendly, and we've switched from TensorFlow to PyTorch for more reliable deployments.

Warning

This major version release breaks all backwards compatibility with previous versions. If you used the DeepLC CLI or Python API in your applications, checkout the migration guide to how to update your code.

Added

  • Multitask pretrained model as the new default, trained across multiple LC setups; automatic head selection in calibrate() based on Pearson correlation
  • predict_and_calibrate(), finetune_and_predict() core functions
  • Automatic calibration reference selection from input PSMs using q-value filtering or top-scoring fraction
  • Calibration.selected_model_head field
  • Built-in transfer learning via adapter-based fine-tuning (replaces deeplcretrainer)
  • NiceGUI web interface (deeplc gui / deeplc gui --native)
  • [gui] and [web] optional dependency groups
  • Docker image for containerized web server deployment
  • Windows one-click installer (PyInstaller + Inno Setup)
  • Sphinx-based documentation on ReadTheDocs
  • CI publish workflow with Windows installer and Docker image builds

Changed

  • PyTorch replaces TensorFlow as the deep learning backend
  • Class-based DeepLC API replaced by standalone functions (predict, calibrate, finetune, train, save_model, etc.)
  • Calibration split into a dedicated reusable module with sklearn-like API
  • CLI restructured into predict and gui subcommands
  • Input format uses psm_utils — accepts Sage, MaxQuant, mzTab, and others; peptide sequences in ProForma 2.0 notation
  • Removed ensemble prediction (three kernel sizes averaged); single model used
  • Improved spline calibration efficiency and default parameters
  • Modernized CI workflows to use uv

Removed

  • Library feature for storing past predictions
  • Legacy CALLC functionality
Assets 3
Loading