ShareNet is a Bayesian framework for boosting the accuracy of cell type-specific gene regulatory associations that can be used with a wide range of general network inference algorithms. It takes as input a set of initial network estimates for each cell type in a dataset, computed using a user's network inference algorithm of choice, and then outputs a corresponding set of revised networks.
Here is an example workflow for using ShareNet with the outputs of a chosen network inference algorithm.
# list of network edge scores
# (each element in the list corresponds to the matrix of network edge scores for a unique cell
# type - e.g., T x R matrix for a network considering T target genes, R regulator genes)
# Note: networks must be of the same size
X = [list of numpy.ndarray]
# list of network edge score standard deviations
# (each element in the list corresponds to the standard deviations of
# the network edge scores for a unique cell type)
V = [list of numpy.ndarray]
import sharenet
# n_components = number of mixture components to use
model = sharenet.ShareNet(n_components = n_components)
# update model parameters using CAVI
model.fit(X,V)
# get revised network edge scores
model.get_revised_edge_scores()Here is an example using the Bayesian variable selection generative model.
# dictionary of cell type-specific gene expression profiles, where keys are the
# names/numbers associated with a cell type and the values are arrays of the corresponding
# gene expression data (the columns in each array should correspond to the same set of
# genes across the various cell types)
celltype_data_dict = {1: numpy.ndarray ,..., C: numpy.ndarray}
import sharenet_bvs as snb
model = snb.ShareNetGeneModel(celltype_data_dict,use_sharenet=True)
# update model parameters (n_processes specifies number of processes for parallelization)
model.fit(n_processes = n_processes)
# get model parameters
model.get_model_params(param_name, celltype_no, target_ind)A couple of examples using real single-cell datasets can be found in sharenet/example1 and sharenet/example2. Example 1 provides a step-by-step walkthrough of a smaller example that uses data from the BEELINE benchmarking dataset. Example 2 includes a comprehensive example related to the mouse blood lineage dataset, in which ShareNet is applied to networks inferred with PIDC, and the accuracy of the networks inferred with and without ShareNet are compared. Jupyter notebooks are provided in each of the corresponding subdirectories with more detailed descriptions.
For questions about the code, contact alexwu@mit.edu.