CelLink.model.Cellink.iterative_unbalanced_ot
- Cellink.iterative_unbalanced_ot(arr1, arr2, ori_cell_id, bot_match_result, reg=5e-3, reg_m=(2, 0.1), numItermax=1000, sparse=False, direction=1)[source]
stage II of celLink: perform iterative unbalanced optimal transport to correct alignments, filter out the matched cells and retain the unmatched cells.
- Parameters:
arr1 (np.array) – The shared-feature arr1
arr2 (np.array) – The shared-feature arr2
ori_cell_id (list) – Cell ids for unmatched cells in phase I
reg (float, default = 0.05) – The penalty coefficient of entropy regularization of UOT
reg_m (tuple) – Each element in the tuple is the penalty coefficient of the KL divergence of the transported mass from one modality and the initial mass from another
numItermax (int) – Max number of iterations of sinkhorn algorithms
direction (int, default = 1) – The direction to align between modalities. If direction is 1, align arr1 -> arr2. If direction is 2, align arr2 -> arr1.
- Returns:
match_result (np.array) – Keys are the cell-cell correspondence matrix fo all matched cells and values are the corresponding vectors.
ori_cell_id (list,) – Cell ids for the unmatched cells after convergence