CBP-SMF an improved Semi-supervised Matrix tri-Factorization framework for characterizing Complex Biological Processes that represent sample groups

This software package contains all the functions of CBP-SMF. Users can use the command line 'pip install CBP_SMF' to install or download CBP_SMF.py and import it.

CBP-SMF factorization decomposes several non-negative matrix X_i into three matrices: Molecular Coefficient Matrix U_i, Factor Absorbing W_i, Sample Basis Matrix V. We use Euclidean distance as cost function(so it's an optimization problem) to measure the distance between X_i and reconstructed U_iW_iV. Besides, we incorporate samples’ label information into NMF through a graph embedding constraint and we give different input X_i a unique weight to donate its contribution in optimization. Given input matrices X_i, labeled samples' subgroup, and a correlation matrix of labeled samples, CBP-SMF integrate MG data (e.g., copy number variation, gene expression, microRNA expression, and/or gene network) to classify the unlabeled samples into groups and identify the underlying CBPs which characterize functional properties of each group.