Framework for running and analyzing ENCODE DHS presence/absence data with non-negative matrix factorization (NMF).
This OONMF.py library provides a wrapper to run the scikit-learn NMF routine, and several other functions that can analyze the resulting decomposed matrices. OONMFhelpers.py and OONMFmetadata.py are libraries that provide additional functions useful for analysis.
To run the code as we did in the DHS Vocabulary paper with k=16, download the repository, as well as the DHS presence/absence matrix (733 samples x 3.59e6 DHSs) dat_bin_FDR01_hg38.txt.gz (available here) and run:
python OONMF_compute_presence_NNDSVD_O.py
the output of this code will be the Basis (733 samples x 16 components) and Mixture (3.59e6 DHSs x 16 components) matrices, in numpy binary format. If tab-separated output format is desired, uncomment the last line of this script.
Note: requires Python 3 with scikit-learn, numpy, scipy, matplotlib, pandas installations.