scself

Self Supervised Tools for Single Cell Data

Molecular Cross-Validation for PCs arXiv manuscript

mcv(
    count_data,
    n=1,
    n_pcs=100,
    random_seed=800,
    p=0.5,
    metric='mse',
    standardization_method='log',
    metric_kwargs={},
    silent=False,
    verbose=None,
    zero_center=False
)

Noise2Self for kNN selection arXiv manuscript

noise2self(
    count_data,
    neighbors=None,
    npcs=None,
    metric='euclidean',
    loss='mse',
    loss_kwargs={},
    return_errors=False,
    connectivity=False,
    standardization_method='log',
    pc_data=None,
    chunk_size=10000,
    verbose=None
)

Implemented as in DEWÄKSS

Feature module and submodule determination using pearson correlation distance, kNN embedding, and leiden clustering

get_correlation_modules(
    adata,
    layer='X',
    n_neighbors=10,
    leiden_kwargs={},
    output_key='gene_module',
    obs_mask=None

)

get_correlation_submodules(
    adata,
    layer='X',
    n_neighbors=10,
    leiden_kwargs={},
    input_key='gene_module',
    output_key='gene_submodule',
    obs_mask=None
)

Feature module and submodule scoring

score_all_modules(
    adata,
    modules=None,
    module_column='gene_module',
    output_key_suffix='_score',
    obs_mask=None,
    layer='X',
    scaler=TruncMinMaxScaler(),
    fit_scaler=True,
    clipping=None
)

score_all_submodules(
    adata,
    modules=None,
    submodules=None,
    module_column='gene_module',
    submodule_column='gene_submodule',
    output_key_suffix='_score',
    obs_mask=None,
    layer='X',
    scaler=TruncMinMaxScaler(),
    fit_scaler=True,
    clipping=None
)