structboost.extract_gene_rankings

structboost.extract_gene_rankings(adata_path, model_key='BAE_encoder_weights', dimensions=None, top_k=50)[source]

Extract ranked gene lists (pos/neg) per latent dimension from .h5ad.

For each dimension, genes are split by sign of their encoder weight. Within each group, genes are ranked by magnitude (descending). At most min(top_k, n_available) genes are returned per group.

Parameters:

  • adata_path (str | Path) – Path to a .h5ad file containing fitted BAE results.

  • model_key (str) – Key in adata.varm holding the encoder weight matrix, e.g. "BAE_encoder_weights".

  • dimensions (list[int] | None) – Indices of latent dimensions to extract. None extracts all.

  • top_k (int) – Maximum number of genes per sign group (positive/negative). Clamped to min(top_k, n_available) when fewer genes exist.

Returns:

list[DimensionGeneRanking] – Ranked gene lists for each requested dimension.

Raises:
Return type:

list[DimensionGeneRanking]