vanheeringen-lab / psap Goto Github PK

CLI interface for the PSAP classifier. PSAP implements a RandomForest approach to predict the probability of proteins to mediate protein phase separation (PPS). Initially, a set of protein sequences is annotated with biochemical features which are subsequently used to train a RandomForest (scikit-learn) classifier. The trained classifier is afterwards exported to json format and can be used to predict the llps class probability (PSAP_score) for unseen protein sequences.

The default model was trained on the human reference proteome with a list of literature curated PPS proteins for positive class labeling. Both can be found in /data.

Publication | Mierlo, G., Jansen, J. R. G., Wang, J., Poser, I., van Heeringen, S. J., & Vermeulen, M. (2021). Predicting protein condensate formation using machine learning. Cell Reports, 34(5), 108705. https://doi.org/10.1016/j.celrep.2021.108705.

• Free software: MIT license

pip

pip install psap

conda

conda install bioconda::psap

psap train -f /path/to/peptide-trainingset.fasta -l /path/to/known/pps-proteins.txt (optional) -o /path/to/output/directory (optional)

The trained RandomForest classifier will be exported to json format and stored in the output directory.

psap predict -f /path/to/peptide-testset.fasta -m /path/to/model.json (optional) -o /path/to/output/directory (optional)

psap loads the default classifier stored in /data/model when no model is provided with -m. An example peptide fasta file can be found in the psap/data/testset folder.

psap annotate -f /path/to/peptide.fasta -l /path/to/known/pps-proteins.txt (optional) -o /path/to/output/directory (optional)

Manually annotate a peptide fasta with biochemical features. This step is included in train and predict.

This package was adapted from the cookiecutter and the audreyr/cookiecutter-pypackage project template.