xchrom.tl.crosscell_nsls

xchrom.tl.crosscell_nsls(cell_embedding_ad: str | Path | AnnData, input_folder: str | Path = './train_data', model_path: str | Path = './train_out/E1000best_model.h5', cal_preddata: bool = True, output_path: str | Path = './eval_out', cellembed_raw: str = 'X_pca', celltype: str = 'celltype', save_pred: bool = False, plot_umap: bool = False)[source]

Evaluate the performance in cross-cell prediction with within-sample data, calculate neighbor score and label score for test cells. Predict all data (excluding cross-peak peaks), then extract test cells to calculate nsls

Parameters:

  • cell_embedding_ad (str or Path or anndata.AnnData) – Path to the cell embeddings adata file. provide cell input embeddings for XChrom model prediction.

  • input_folder (str or Path) – Path to the train data folder. Should generate by XChrom_preprocess.py. Must contain ‘splits.h5’, ‘ad_crosscell.h5ad’, ‘m_crosscell.npz’, ‘trainval_seqs.h5’.

  • model_path (str or Path) – Path to the trained model.

  • cal_preddata (bool) – Whether to calculate the nsls score based on the predicted data. If False, calculate nsls score based on the raw atac data.

  • output_path (str or Path) – Path to the output folder.

  • cellembed_raw (str) – Key of the raw cell embeddings in the cell embeddings adata,to calculate RNA neighbors.

  • celltype (str) – Key of the cell type in the cell embeddings adata.

  • save_pred (bool) – Whether to save the prediction matrix with h5ad format.

  • plot_umap (bool) – Whether to plot the UMAP of the test cells.

Returns:

metrics – Dictionary containing neighbor score(k=10,50,100) and label score(k=10,50,100) for test cells.

Return type:

dict

Examples

>>> import xchrom as xc
metrics4 = xc.tl.crosscell_nsls(
    cell_embedding_ad='./data/1_within_sample/m_brain_paired_rna.h5ad',
    input_folder='./data/1_within_sample/train_data',
    model_path='./data/1_within_sample/train_out/E1000best_model.h5',
    output_path='./data/1_within_sample/eval_out',
    cellembed_raw='X_pca',
    celltype='pc32_leiden',
    save_pred=True,
    plot_umap=True
    )