M3JF:Multi-Modal Matrix Joint Factorization for Integrative
Multi-Omics Data Analysis
Multi modality data matrices are factorized conjointly into the multiplication of a shared sub-matrix and multiple
modality specific sub-matrices, group sparse constraint is
applied to the shared sub-matrix to capture the homogeneous and
heterogeneous information, respectively. Then the samples are
classified by clustering the shared sub-matrix with kmeanspp(),
a new version of kmeans() developed here to obtain concordant
results. The package also provides the cluster number
estimation by rotation cost. Moreover, cluster specific
features could be retrieved using hypergeometric tests.
