def pcr_noise(
read_counts: np.ndarray,
pcr_betas: Union[float, np.ndarray],
n_cycles: int,
copy: bool = True,
) -> np.ndarray:
"""PCR noise model: every read has an affinity for PCR, and for every round
of PCR we do a ~binomial doubling of each count.
:param read_counts: array of shape (n_samples, n_features) representing unique
molecules (e.g. genes or gene fragments). If a sparse matrix
is provided, the output will be sparse
:param pcr_betas: PCR efficiency for each feature, either constant or per-feature
:param n_cycles: number of rounds of PCR to simulate
:param copy: if True, return a copy of the read_counts array, else modify in-place
:return: int array of shape (n_samples, n_features) with amplified counts
"""
if np.any(pcr_betas < 0):
raise ValueError("pcr_betas must be non-negative")
if copy:
read_counts = read_counts.copy()
pcr_betas = np.broadcast_to(pcr_betas, (1, read_counts.shape[1]))
if sp.issparse(read_counts):
d = read_counts.data[None, :]
pcr_betas = pcr_betas[:, read_counts.nonzero()[1]]
else:
d = read_counts
# for each round of pcr, each gene increases according to its affinity factor
for i in range(n_cycles):
d += np.random.binomial(n=d, p=pcr_betas, size=d.shape)
if sp.issparse(read_counts):
read_counts.data = d.flatten()
return read_counts
