HoloNet.predicting.mgc_training_for_multiple_targets#

HoloNet.predicting.mgc_training_for_multiple_targets(X, adj, target_all_gene_expr, repeat_num=5, device='cpu', **kwargs)#

Using cell-type tensor and normalized adjancency matrix as the inputs, repeated training GNN to generate the expression of multiple target genes. Also generate the expression only using cell-type information.

Parameters
X Tensor

A tensor (cell_num * cell_type_num) with cell-type information. derived from ‘get_continuous_cell_type_tensor’ or ‘get_one_hot_cell_type_tensor’ function.

adj Tensor

A normalized adjancency matrix derived from ‘adj_normalize’ function.

target_all_gene_expr Tensor

The scaled expression tensor of target genes (cell_num * target_gene_num), from ‘get_gene_expr’ function.

repeat_num int (default: 5)

The number of repeated training, defaultly as 50.

use_gpu

If true, model will be trained in GPU when GPU is available.

kwargs

Other training hyperparameter in ‘mgc_repeat_training’ function.

Return type

Tuple[List[List[MGC_Model]], List[List[MGC_Model]]]

Returns

: Two list of multiple (repeated training) trained MGC model for generating the expression of target genes, one using MGC with cell-type information and CE tensor, one only with cell-type information.