dlomix 0.0.1

Deep Learning for Proteomics

DLOmix

DLOmix is a python framework for Deep Learning in Proteomics. Initially built ontop of TensorFlow/Keras, support for PyTorch can however be integrated once the main API is established.

Usage

Experiment a simple retention time prediction use-case using Google Colab   

Installation

Run the following to install:

$ pip install dlomix

Package Structure:

• data.py: structures for modelling the input data, currently based on tf.Dataset
• models.py: common model architectures for the relevant use-cases based on tf.keras.Model to allow for using the Keras training API
• pipeline.py: an exemplary high-level pipeline implementation
• eval.py: classes for evaluating models and reporting results
• eval_utils.py: custom evaluation metrics implemented in TensorFlow/Keras
• constants.py: constants and configuration values needs for the pipeline class.

Use-cases:

• Retention Time Prediction (indexed retention time):
  • a regression problem where the the retention time of a peptide sequence is to be predicted.

To-Do:

Functionality:

• integrate prosit
• extend pipeline for different types of models and backbones
• extend pipeline to allow for fine-tuning with custom datasets
• add residual plots to reporting, possibly other regression analysis tools
• output reporting results as PDF
• extend data representation to include modifications

Package structure:

• integrate deeplc.py into models.py, preferably introduce a package structure (e.g. models.retention_time)
• add references for implemented models in the ReadMe
• introduce a style guide and checking (e.g. PEP)
• plan documentation (sphinx and readthedocs)

Developing DLOmix

To install dlomix, along with the the tools needed to develop and run tests, run the following command in your virtualenv:

$ pip install -e .[dev]

References:

[Prosit] Gessulat, S., Schmidt, T., Zolg, D. P., Samaras, P., Schnatbaum, K., Zerweck, J., ... & Wilhelm, M. (2019). Prosit: proteome-wide prediction of peptide tandem mass spectra by deep learning. Nature methods, 16(6), 509-518.

[DeepLC] DeepLC can predict retention times for peptides that carry as-yet unseen modifications Robbin Bouwmeester, Ralf Gabriels, Niels Hulstaert, Lennart Martens, Sven Degroeve bioRxiv 2020.03.28.013003; doi: 10.1101/2020.03.28.013003