| This repo is deprecated! It does not contain the latest updates regarding Prosit! |
| Code related to model training can be found at dlomix! |
| The latest models can be found & accessed on Koina! |
| More functionality (spectral-library generation & rescoring) is available with Oktoberfest! |
| This functionality is also available on the Prosit website! |
Prosit is a deep neural network to predict iRT values and MS2 spectra for given peptide sequences. You can use it at proteomicsdb.org/prosit/ without installation.
Prosit requires
Prosit requires
- Docker 17.05.0-ce
- nvidia-docker 2.0.3 with CUDA 8.0 and CUDNN 6 or later installed
- make 4.1
Prosit was tested on Ubuntu 16.04, CUDA 8.0, CUDNN 6 with Nvidia Tesla K40c and Titan Xp graphic cards with the dependencies above.
The time installation takes is dependent on your download speed (Prosit downloads a 3GB docker container). In our tests installation time is ~5 minutes.
Prosit assumes your models are in directories that look like this:
- model.yml - a saved keras model
- config.yml - a model specifying names of inputs and outputs of the model
- weights file(s) - that follow the template
weights_{epoch}_{loss}.hdf5
You can download pre-trained models for HCD fragmentation prediction and iRT prediction on https://figshare.com/projects/Prosit/35582.
The following command will load your model from /path/to/model/.
In the example GPU device 0 is used for computation. The default PORT is 5000.
make server MODEL_SPECTRA=/path/to/fragmentation_model/ MODEL_IRT=/path/to/irt_model/
Currently two output formats are supported: a MaxQuant style msms.txt not including the iRT value and a generic text file (that works with Spectronaut)
Please find an example input file at example/peptidelist.csv. After starting the server you can run the following commands, depending on what output format you prefer:
curl -F "peptides=@examples/peptidelist.csv" http://127.0.0.1:5000/predict/generic
curl -F "peptides=@examples/peptidelist.csv" http://127.0.0.1:5000/predict/msp
curl -F "peptides=@examples/peptidelist.csv" http://127.0.0.1:5000/predict/msms
The examples take about 4s to run. Expected output files (.generic, .msp and .msms) can be found in examples/.
You can adjust the example above to your own needs. Send any list of (Peptide, Precursor charge, Collision energy) in the format of /example/peptidelist.csv to a running instance of the Prosit server.
Please note: Sequences with amino acid U, O, or X are not supported. Modifications except "M(ox)" are not supported. Each C is treated as Cysteine with carbamidomethylation (fixed modification in MaxQuant).
- Load the models given as in the MODEL_X environment variables
- Start a server and wait for inputs
- On incomming request
- transform peptide list to model input format (numpy arrays)
- predict fragment intensity and iRT with the loaded models for the given peptides
- transform prediction to the requested output format and return response
