omicverse.pp.recover_counts

omicverse.pp.recover_counts(X, mult_value, max_range, log_base=None, chunk_size=1000)[source]

Given log-normalized gene expression data, recover the raw read/UMI counts by inferring the unknown size factors.

Parameters:

  • X –

    The log-normalized expression data. This data is assumed to be normalized via X := log(X/S * mult_value + 1)

    IMPORTANT: This function REQUIRES log-transformed data (e.g., after scanpy.pp.normalize_total() followed by scanpy.pp.log1p()). It will NOT work correctly on linearly normalized data (e.g., data normalized to CPM/TPM without log transformation).

    Expected preprocessing workflow: 1. sc.pp.normalize_total(adata, target_sum=mult_value) 2. sc.pp.log1p(adata) 3. ov.pp.recover_counts(adata.X, mult_value, max_range)

  • max_range – Maximum size-factor search range to use in binary search.

  • mult_value – The multiplicative value used in the normalization. For example, for TPM this value is one million. For logT10K, this value is ten thousand.

  • log_base (default: None) – The base of the logarithm (None for natural log)

Returns:

  • counts – The inferred counts matrix

  • size_factors – The array of inferred size-factors (i.e., total counts)

Raises:

ValueError: – If input data appears to be non-log-transformed (contains very large values that would cause numerical overflow)