A context aware tool for proteasomal cleavage predictions
pepsickle is an open-source command line tool for proteasomal cleavage prediction. pepsickle is designed with flexibility in mind allows for the use of either direct amino acid sequences or the use of FASTA files. Predictions can also be determined based on a variety of available models including those trained on: in-vivo epitope data (default), in-vitro constitutive proteasome data, or in-vitro immunoproteasome data. For information on available models and how they were trained, see the companion paper in Bioinformatics highlighting this tool, as well as the accompanying paper repo with code for training and reproduction.
pepsickle is licensed under the MIT license. See LICENSE for more details.
pepsickle relies on Python 3 and a few other required packages. A complete list of dependencies can be found in requirements.txt
We recommend using a version control system like Anaconda to make sure version requirements for pepsickle don't interfere with other packages in use.
For ease of use, we've provided a .yml file for easy conda setup. After conda is installed, from the main directory simply type:
conda env create --file pepsickle_conda_build.yml
conda activate pepsickle-v0-2-2
This conda environment contains both pepsickle and it's dependencies, which allows use with no other steps required.
If you do not want to use conda, but already have Python 3 installed, pepsickle can simply be installed via the command line by using pip:
pip install pepsickle
pepsickle allows for multiple methods of use. By default, predictions are made based on a model trained using in-vivo epitope data.
During predictions, the upstream and downstream amino acid contexts are used and we therefore recommend including at least 8 amino acids on each side of any sites of interest. If less than the recommended context is given (such as in the case of residues near the beginning or end of a protein sequence) pepsickle will auto-pad inputs. X's submitted to the prediction model are interpreted as the presence of an amino acid sequence with unkonwn identity, while auto-padding * is interpreted as the absence of amino acid context all together.
For predictions on single short amino acid sequences, pepsickle can be run
using the -s option:
pepsickle -s VSGLEQLESIINFEKLTEWTSSNV
For long peptide sequences or to run multiple sequences at once, pepsicklecan be run using the fasta file -f option:
pepsickle -f /PATH/TO/FASTA.fasta
For an example of a FASTA formatted file, see the test fasta used for this package.
By default, output will be printed to the screen, however output can easily be routed to a file location by using the -o option:
pepsickle -s VSGLEQLESIINFEKLTEWTSSNV -o /PATH/TO/OUTPUT.txt
Output is in tab separated format. For an example of output format see the example out file.
A full list of command line options and descriptions is listed here:
-s, --sequence [SEQUENCE] use pepsickle in single sequence mode. Takes a string sequence as input and returns predicted cleavage sites in standard format.
-f, --fasta [FASTA] use pepsickle in fasta mode. Takes a fasta file with protein ID's and corresponding sequences.
-o, --out [OUT_FILE] name and destination for prediction outputs in TSV format. If none is provided, the output will be printed directly to the terminal.
-v, --verbose In fasta mode, prints progress during cleavage predictions for fasta files with > 100 protein sequences.
-m, --model-type [epitope (default) | in-vitro | in-vitro-2] allows the use of models trained on alternative types of data. Defaults to epitope based model, with options for in-vitro based gradient boosted model (in-vitro) or an experimental neural network based in-vitro model (in-vitro-2).
-p, "--proteasome-type [C | I] allows predictions to be made based on constitutive proteasomal (C) or immunoproteasomal (I) cleavage profiles. Note that if predictions are made using the epitope-based model (default), predictions will be proteasome type agnostic.
-t, --threshold [0-1 (default=0.5)] probability threshold to be used for cleavage predictions.
--human-only (experimental) uses models trained on human data only. Note that human only data sets are substantially smaller and may produce less stable predictions.