Celluloid: clustering single cell sequencing data around centroids
pip3 install celluloid-clustSingle Cell file
The input file (specified by the -i parameter) is expected to be a ternary matrix file where the rows represent the cells and the columns the mutations. Each cell must be separated by a space. Each cell of the matrix can be:
| Value of cell | Meaning |
|---|---|
| I[i,j] = 0 | Mutation j is not observed in cell i |
| I[i,j] = 1 | Mutation j is observed in cell i |
| I[i,j] = 2 | There is no information for mutation j in cell i, i.e. low coverage |
Mutations file
This optional file specifies the name of the mutations (parameter -l). Each mutation's name must be on a different line (separated by \n), and the names are assigned to columns from left to right in the input file. If this file is not provided, mutations are progressively named from 1 to the total number of mutations.
Celluloid will output one Single Cell file and one Mutations file in the same format as the input but with a number of mutations reduced by the clustering. Labels of mutation are concatenated in the out label file.
usage: celluloid.py [-h] -i INPUT [-m {huang,random}] [-d {conflict,matching}]
[-n N] -k K [-l LABELS] -o OUTDIR [-v]
Celluloid: clustering single cell sequencing data around centroids
optional arguments:
-h, --help show this help message and exit
-i INPUT, --input INPUT
input file
-m {huang,random}, --method {huang,random}
initialization method
-d {conflict,matching}, --dissim {conflict,matching}
dissimilarity measure
-n N, --n_inits N number of iterations
-k K, --kmodes K number of modes
-l LABELS, --labels LABELS
label file
-o OUTDIR, --outdir OUTDIR
output directory.
-v, --verbose verbose
The preprint of an article outlining a study on Celluloid can be found at https://doi.org/10.1101/586545
A workflow for reproducing the experiments of this study is described as a "Snakefile", which needs to be run with snakemake.
Overview
The workflow is designed to be run from this directory. In principle,
you only need to run snakemake here and then the rest is done
automatically, but all dependencies need to be installed first. To
install dependencies, see How to install, and then
install sasc:
git clone https://github.com/sciccolella/sasc
cd sasc && gcc -o sasc *.c -std=gnu99 -g -lm
Run the workflow
Start the workflow with
nice snakemake -p -j 16
Adjust the 16, which is the number of cores you want to use.
Results
The resulting files for each dataset, e.g.,
data/exp1/sim_1_scs.txt will be reported in a directory named
data/exp1/sim_1.{parameters}.celluloid (resp.,
data/exp1/sim_1.{parameters}.kmeans) for the clustering -- and the
downstream run of sasc -- with celluloid (resp., kmeans), where
{parameters} are the parameters, e.g., initialization method,
dissimilarity measure, k, associated with corresponding clustering
method