Heimdall.fg.Fg.preprocess_embeddings

abstractmethod Fg.preprocess_embeddings(float_dtype='float32')

Preprocess gene embeddings and store them for use during model inference.

Preprocessing may include anything from downloading gene embeddings from a URL to generating embeddings from scratch.

Parameters:

float_dtype (str) – dtype to be used for identity embedding state.

Returns:

Sets self.gene_embeddings. Sets the following fields of self.adata: .var[‘identity_embedding_index’] : ndarray (shape (self.adata.n_vars,))

Index of gene in embeddings.

.var[‘identity_valid_mask’]ndarray (shape (self.adata.n_vars,))

Boolean mask indicating whether or not gene is mapped by this Fg.