scSHARP Model

GCN Model

GCN model class propagates labels from component tool predictions. For more detailed model information see paper

class scSHARP.sc_sharp.GCNModel(config_file, neighbors, target_types, seed=8, dropout=0.0)

class for label propagation GCN model

construct_graph(X)

construct graph from data

forward(X, edge_index=None)

forward pass through model

predict(testloader)

predict with model

to(device)

send layers to device

train(trainloader, epochs, verbose=True, ignore_index_input=-1)

train model

validation_metrics(testloader, train_nodes, test_nodes)

returns validation metrics (ROC, prC, accuracy, etc) test_nodes is array with indices of nodes whose labels were not used for training

PCA Model

class scSHARP.sc_sharp.PCAModel(pca_comps, pca_mean)

class for pca model that takes in pre fitted components

forward(X)

forward pass for pca transformation