omicverse.single.cNMF

omicverse.single.cNMF#

class omicverse.single.cNMF(adata, components, n_iter=100, densify=False, tpm_fn=None, seed=None, beta_loss='frobenius', num_highvar_genes=2000, genes_file=None, alpha_usage=0.0, alpha_spectra=0.0, init='random', output_dir=None, name=None, use_gpu=True, gpu_id=0)[source]#

Consensus NMF workflow wrapper for robust gene-program discovery.

Parameters:

  • adata (AnnData) – Input single-cell expression AnnData.

  • components (array-like) – Candidate rank values (number of programs/topics) to evaluate.

  • n_iter (int, default=100) – Number of NMF restarts per rank.

  • output_dir (str, optional) – Directory used to store temporary and result files.

  • name (str, optional) – Analysis name prefix for output artifacts.

  • use_gpu (bool, default=True) – Whether to use GPU-accelerated factorization when available.

  • gpu_id (int, default=0) – CUDA device index.

__init__(adata, components, n_iter=100, densify=False, tpm_fn=None, seed=None, beta_loss='frobenius', num_highvar_genes=2000, genes_file=None, alpha_usage=0.0, alpha_spectra=0.0, init='random', output_dir=None, name=None, use_gpu=True, gpu_id=0)[source]#
Parameters:

  • output_dir (default: None) – path, optional (default=None). Output directory for analysis files. If None, all analysis is done in memory.

  • name (default: None) – string, optional (default=None). A name for this analysis. Will be prefixed to all output files. If set to None, will be automatically generated from date (and random string).

  • use_gpu (default: True) – bool, optional (default=True). If True and GPU is available, use GPU acceleration for NMF factorization.

  • gpu_id (default: 0) – int, optional (default=0). GPU device ID to use when multiple GPUs are available.

Methods

__init__(adata, components[, n_iter, ...])

calculate_silhouette_k(k[, ...])

Calculate silhouette scores for a given k value.

combine([components, skip_missing_files])

Combine NMF iterations for the same value of K :type components: default: None :param components: Values of K to combine iterations for.

combine_nmf(k[, skip_missing_files, ...])

consensus(k[, density_threshold, ...])

Obtain consensus estimates of spectra and usages from a cNMF run and output a clustergram of the consensus matrix.

factorize([worker_i, total_workers])

Iteratively run NMF with prespecified parameters.

factorize_multi_process(total_workers)

multiproces wrapper for nmf.factorize() factorize_multi_process() is direct wrapper around factorize to be able to launch it form mp.

get_nmf_iter_params(ks[, n_iter, ...])

Create a DataFrame with parameters for NMF iterations.

get_norm_counts(counts, tpm[, ...])

get_results(adata, result_dict)

get_results_rfc(adata, result_dict[, ...])

Map cNMF results to adata and train RFC/decision-tree cluster labels.

k_selection_plot([close_fig, ...])

Plot stability, reconstruction error, and optionally silhouette scores for K selection.

load(filename)

Load a saved cNMF object from file.

load_results(K, density_threshold[, ...])

Loads normalized usages and gene_spectra_scores for a given choice of K and local_density_threshold for the cNMF run.

plot_silhouette_for_k(k[, ...])

Plot detailed silhouette plot for a single k value.

plot_silhouette_survey([k_range, ...])

Generate a grid of silhouette plots for multiple k values for easy comparison.

prepare(adata, components[, n_iter, ...])

Load input counts, reduce to high-variance genes, and variance normalize genes.

refit_spectra(X, usage)

Takes an input data matrix and a fixed usage matrix and uses NNLS to find the optimal spectra matrix.

refit_usage(X, spectra)

Takes an input data matrix and a fixed spectra and uses NNLS to find the optimal usage matrix.

save(filename)

Save the cNMF object to a file for later use.

save_nmf_iter_params(replicate_params, ...)

save_norm_counts(norm_counts)

Contents