scSLAT.model.utils.spatial_match

scSLAT.model.utils.spatial_match(embds, reorder=True, smooth=True, smooth_range=20, scale_coord=True, adatas=None, return_euclid=False, verbose=False, get_null_distri=False)

Use embedding to match cells from different datasets based on cosine similarity

Parameters:

• embds (typing.List[torch.Tensor]) – list of embeddings

• reorder (typing.Optional[bool]) – if reorder embedding by cell numbers

• smooth (typing.Optional[bool]) – if smooth the mapping by Euclid distance

• smooth_range (typing.Optional[int]) – use how many candidates to do smooth

• scale_coord (typing.Optional[bool]) – if scale the coordinate to [0,1]

• adatas (typing.Optional[typing.List[anndata._core.anndata.AnnData]]) – list of adata object

• verbose (typing.Optional[bool]) – if print log

• get_null_distri (typing.Optional[bool]) – if get null distribution of cosine similarity

Note

Automatically use larger dataset as source

Return type:

typing.List[typing.Union[torch.Tensor, numpy.ndarray]]

Returns:

Best matching, Top n matching and cosine similarity matrix of top n

Note

Use faiss to accelerate, refer https://github.com/facebookresearch/faiss/issues/95