This tool uses non-negative matrix factorization to impute missing values in quantitative mass spectrometry data.
With the python standard library venv module
python3 -m venv ms_imputer
source ms_imputer/bin/activate
pip3 install -e .
With conda
conda create -n ms_imputer python=3.7
conda activate ms_imputer
pip3 install -e .
Usage: ms_imputer [OPTIONS]
Fit an NMF model to the input matrix, impute missing values.
Options:
--csv_path TEXT path to the input matrix (.csv) [required]
--output_stem TEXT file stem to use for output file [required]
--factors INTEGER number of factors to use for reconstruction
--learning_rate FLOAT the optimizer learning rate
--max_epochs INTEGER max number of training epochs
--help Show this message and exit.
The factors, learning_rate and max_epochs params are not required.
They correspond to the number of latent factors to use in training the NMF model, the initial learning rate for the Adam optimizer and the maximum number of training iterations for the model, respectively.
Unless otherwise specified, these will be set to:
factors : 8
learning_rate : 0.05
max_epochs : 3000
This work was produced by Lincoln Harris, Bill Noble and Seewong Oh, of the University of Washington, and Will Fondrie of Talus Bioscience. For questions please contact lincolnh@uw.edu.
We welcome any bug reports, feature requests or other contributions. Please submit a well documented report on our issue tracker. For substantial changes please fork this repo and submit a pull request for review.
See CONTRIBUTING.md for additional details.
You can find official releases here.