mtopic.pp.feature_associations_data

mtopic.pp.feature_associations_data(mdata, mod_list, normalize=True)

Prepare feature signatures for training feature associations.

Extracts and optionally row-normalizes signature matrices from multiple modalities in a MuData object. Each signature matrix is expected to be stored in varm[‘signatures’] of the respective modality. Matrices are returned in a dictionary keyed by modality name.

Parameters:

• mdata (muon.MuData) – A MuData object containing modalities with signature matrices in varm[‘signatures’].

• mod_list (list[str]) – List of modality names to extract from mdata.

• normalize (bool) – If True, rows of the signature matrices are normalized to sum to 1. (default: True)

Returns:

A dictionary where keys are modality names and values are tuples (M, M_var), with: - M: torch.FloatTensor of shape (n_topics, n_features), the signature matrix. - M_var: The corresponding var DataFrame from the modality.

Return type:

dict[str, tuple[torch.Tensor, pandas.DataFrame]]

Example:

import muon as mu
import torch

mdata = mu.read('multiome_data.h5mu')

# Extract signatures from RNA and ATAC modalities
associations_input = feature_associations_data(mdata, ['rna', 'atac'])

A, A_var = associations_input['rna']
B, B_var = associations_input['atac']