/DecOT

Primary LanguagePython

DecOT

DecOT is a bulk gene expression deconvolution method that uses the optimal transport distance as loss and apply an ensemble framework to integrate reference information from scRNA-seq data of multiple individuals.

Installation

DecOT is a python script developed in a Windows and Linux 64-bit architecture. To use DecOT, the following packages need to be installed：

python packages:

  • Pandas
  • Numpy
  • Scipy
  • POT
  • rpy2
  • multiprocess
  • matlab

R packages:

  • WGCNA

DecOT uses wasserstein_NMF_coefficient_step.m developed by Rolet(2016) to calculate Wasserstein loss. We rewrite the .m files into .py format, which is suitable for small batches of data. For data with thousands of genes, we recommend using DecOT with matlab interface.

  • Rolet, A., Cuturi, M., & Peyré, G. (2016, May). Fast dictionary learning with a smoothed Wasserstein loss. In Artificial Intelligence and Statistics (pp. 630-638). PMLR.

Input file

The input files are tab-delimited .txt or .csv files, which contain:

  • Y.csv : Bulk mixtures (genes * mixtures)
  • C.csv : Reference cells (genes * cells)
  • pDataC.csv : Cell phenotype data with three columns ('cellID', 'cellType', 'sampleID')

Parameters

  def decot_ensemble(Y, C, pDataC, ground_cost_path, metric, save_path, cores=4, gamma=0.001, rho=0.001):
    Y: `pandas dataframe`
        Bulk mixtures (genes x mixtures).

    C: `pandas dataframe`
        Reference cells (genes x cells).

    pDataC: `pandas dataframe`
        Cell phenotype data (cells x ['cellID', 'cellType', 'sampleID'])

    ground_cost_path: str
        Ground cost matrix file path.

    metric: str
        Metric used to compute the ground cost (['dissTOM', 'euclidean', 'cosine', 'correlation']).

    save_path: str
        Path to save deconvolution results.

    cores: int
        Number of cores used for parallel computing. (for ensemble framework)

    gamma: regularization parameter
        Default: 0.001

    rho: regularization parameter
        Default: 0.001

Example

An example dataset (simulation/Batch1) is provided to run DecOT with default parameters (gamma: 0.001, rho: 0.001). To run DecOT, please follow commands below from the DecOT directory.

  • python DecOT.py --working_dir ../results/simulation/Batch1/ --y sim0_Y.csv --c sim0_C.csv --pDataC sim0_pDataC.csv --metric dissTOM --gamma 0.001 --rho 0.001
  • python DecOT.py --ensemble --working_dir ../results/simulation/Batch1/ --y sim0_Y.csv --c sim0_C.csv --pDataC sim0_pDataC.csv --metric dissTOM --cores 6 --gamma 0.001 --rho 0.001

We also provide a toy dataset which are mixtures of gauss distributions:

  • python DecOT.py --random_gauss_test --genes 100 --types 5 --mixtures 10

You can download our source code and use DecOT in command line or install the DecOT package and directly call the functions in it:

    python setup.py install