AtlasMapper

hnoca.map.mapper.AtlasMapper(ref_model)

A class to map a query dataset to a reference dataset using scPoli, scVI or scANVI models.

Parameters:

Name	Type	Description	Default
ref_model	Union[SCANVI, SCVI, scPoli]	The reference model to map the query dataset to.	required

map_query(query_adata, retrain='partial', labeled_indices=None, **kwargs)

Map a query dataset to the reference dataset

Parameters:

Name	Type	Description	Default
query_adata	AnnData	The query dataset to map to the reference dataset	required
retrain	Literal['partial', 'full', 'none']	Whether to retrain the query model. "partial" will retrain the weights of the new batch key "full" will retrain the entire model "none" will use the reference model without retraining	'partial'
labeled_indices	Optional[ndarray]	The indices of labeled cells in the query dataset. This is only used for scPoli models.	None
**kwargs		Additional keyword arguments to pass to the training function	{}

compute_wknn(ref_adata=None, k=100, query2ref=True, ref2query=False, ref_rep_key='X_latent', query_rep_key='X_latent', weighting_scheme='jaccard_square', top_n=None)

Compute the weighted k-nearest neighbors graph between the reference and query datasets

Parameters:

Name	Type	Description	Default
k	int	Number of neighbors per cell	100
query2ref	bool	Consider query-to-ref neighbors	True
ref2query	bool	Consider ref-to-query neighbors	False
weighting_scheme	Literal['n', 'top_n', 'jaccard', 'jaccard_square', 'gaussian', 'dist']	How to weight edges in the ref-query neighbor graph	'jaccard_square'
top_n	Optional[int]	The number of top neighbors to consider	None

get_presence_scores(split_by=None, random_walk=True, alpha=0.1, n_rounds=100, log=True)

Estimate the presence score of the query dataset

Parameters:

Name	Type	Description	Default
split_by	str	The column in the query dataset to split by	None
random_walk	bool	Whether to use random walk to estimate presence score	True
alpha	float	The heat diffusion parameter for the random walk	0.1
n_rounds	int	The number of rounds for the random walk	100
log	bool	Whether to log the presence score	True

Returns:

Type	Description
dict	A dictionary with the presence scores

transfer_labels(label_key)

Transfer labels from the reference dataset to the query dataset

Parameters:

Name	Type	Description	Default
label_key	str	str The column in the reference dataset to transfer	required

Returns:

Type	Description
dict	A dictionary with the transfer scores

get_matched_expression(rescale_factor=1)

Get the expression of reference cells matched to query cells. This can be used for quantitative comparisons like DE analysis.

Parameters:

Name	Type	Description	Default
rescale_factor	int	str Factor to rescale the log-normalized counts	1

Returns:

Type	Description
AnnData	An AnnData object with the matched expression

save(output_dir)

Save the mapper object to disk

Parameters:

Name	Type	Description	Default
output_dir	str	str The directory to save the mapper object	required

load(input_dir) classmethod

Load the mapper object from disk

Parameters:

Name	Type	Description	Default
input_dir	str	str The directory to load the mapper object	required