cellink.tl.external.JointNMFWrapper#

class cellink.tl.external.JointNMFWrapper(Xh, Xd, *, n_shared=10, n_healthy_specific=5, n_disease_specific=5, gamma=1.0, mu=None, n_init=5, max_iters=1000, tol=1e-10, random_state=0)#

Joint NMF decomposition for paired healthy / disease single-cell data.

Parameters:

Xh (ndarray) – Dense matrix (cells x genes) for the healthy condition.
Xd (ndarray) – Dense matrix (cells x genes) for the disease condition. Must have the same shape as Xh (paired cells x genes) since factor alignment correlates loadings across the two conditions cell-by-cell.
n_shared (int (default: 10)) – Number of programs shared between healthy and disease (KC in paper).
n_healthy_specific (int (default: 5)) – Number of programs unique to the healthy condition (KH in paper).
n_disease_specific (int (default: 5)) – Number of programs unique to the disease condition (KD in paper).
gamma (float (default: 1.0)) – Coupling strength. Higher values force shared programs to align more.
mu (float | None (default: None)) – L2 regularisation on loading matrices. If None, estimated from data.
n_init (int (default: 5)) – Number of random NMF initialisations; best reconstruction is kept.
max_iters (int (default: 1000)) – Maximum multiplicative-update iterations.
tol (float (default: 1e-10)) – Convergence tolerance on relative change in cost.
random_state (int (default: 0)) – Random seed for reproducibility.

Attributes table#

`Hd`
`Hh`
`Wd`
`Wh`	Healthy cell x factor loadings (dense).

Methods table#

fit()

Run multiplicative updates until convergence.

Attributes#

JointNMFWrapper.Hd#

JointNMFWrapper.Hh#

JointNMFWrapper.Wd#

JointNMFWrapper.Wh#: Healthy cell x factor loadings (dense).

Methods#

JointNMFWrapper.fit()#