Result types

Frozen dataclasses over numpy arrays.

Frozen result records.

Plain dataclasses over numpy arrays. eq=False because elementwise == on arrays does not return a bool, so a generated __eq__ would raise.

Tree dataclass

Tree(parent: ndarray, branch: ndarray, n_leaves: int)

A rooted tree over the cells.

Leaves are 0..n_leaves in input order; internal nodes follow, ordered by height above the leaves. Every function in this package that takes a tree assumes that ordering, so build trees with this package (or read_newick) rather than by hand.

Attributes:

Name	Type	Description
parent	ndarray	Parent of each node, -1 for the root. int64.
branch	ndarray	Length of the branch above each node; the root's is ignored.
n_leaves	int	Number of leaves.

n_nodes property

n_nodes: int

Leaves plus inferred ancestors.

BonsaiResult dataclass

BonsaiResult(
    tree: Tree,
    loglik: float,
    features: ndarray,
    dropped: ndarray,
    node_means: ndarray,
    node_sds: ndarray,
    steps: list[Step],
)

A finished reconstruction.

Attributes:

Name	Type	Description
tree	Tree	The tree. Leaf i is input cell i.
loglik	float	Final loglikelihood, meaningful only up to an additive constant.
features	ndarray	Input column of each retained feature, ascending. For bonsai_from_counts these index the genes of the count matrix.
dropped	ndarray	Genes the Sanity conversion dropped as ill-conditioned. Empty unless the run started from counts.
node_means	ndarray	Posterior mean of every node, (n_nodes, n_retained), in the input's units. Rows n_leaves: are the ancestors.
node_sds	ndarray	Posterior standard deviation, same layout.
steps	list[Step]	The loglikelihood after each search step, in order.

Step dataclass

Step(step: str, loglik: float, gain: float)

What one step of the search bought.

Attributes:

Name	Type	Description
step	str	Which step, "1 star" through "8 collapse".
loglik	float	Tree loglikelihood after it, up to an additive constant.
gain	float	Change from the previous step.

SanityResult dataclass

SanityResult(
    log_fold_changes: ndarray,
    error_bars: ndarray,
    mean_log_quotient: ndarray,
    mean_log_quotient_error: ndarray,
    variance: ndarray,
)

Sanity posteriors.

Attributes:

Name	Type	Description
log_fold_changes	ndarray	Posterior log fold change d_c, (n_cells, n_genes). Zero-centred per gene. This, not log_transcription_quotients, is what from_sanity wants.
error_bars	ndarray	Posterior SD on each log fold change, same layout.
mean_log_quotient	ndarray	Per-gene mean log transcription quotient m.
mean_log_quotient_error	ndarray	Error bar on m.
variance	ndarray	Per-gene variance of the log fold changes v.

log_transcription_quotients property

log_transcription_quotients: ndarray

Normalised expression m + d_c, (n_cells, n_genes).

Likelihood dataclass

Likelihood(
    means: ndarray,
    sds: ndarray,
    variances: ndarray,
    features: ndarray,
    dropped: ndarray,
)

Likelihood means and SDs recovered from Sanity posteriors (S5).

Attributes:

Name	Type	Description
means	ndarray	(n_cells, n_kept), ready for bonsai.
sds	ndarray	Same layout.
variances	ndarray	Sanity's v for the kept genes; pass it to bonsai.
features	ndarray	Input column of each kept gene.
dropped	ndarray	Input columns dropped as ill-conditioned.

Clustering dataclass

Clustering(
    leaf_cluster: ndarray, centres: ndarray, sizes: ndarray
)

A cut of the tree into clusters.

Attributes:

Name	Type	Description
leaf_cluster	ndarray	Cluster of each leaf, numbered by decreasing size.
centres	ndarray	Representative node of each cluster.
sizes	ndarray	Leaves per cluster.

SimulatedData dataclass

SimulatedData(
    tree: Tree,
    truth: ndarray,
    means: ndarray,
    sds: ndarray,
    variances: ndarray,
)

Brownian motion on a known tree.

Attributes:

Name	Type	Description
tree	Tree	The generating tree.
truth	ndarray	Noise-free leaf positions, (n_leaves, n_features).
means	ndarray	Observed means, truth plus noise.
sds	ndarray	Error bars on the means.
variances	ndarray	Per-feature variance the data was scaled by. truth, means and sds are already divided by its square root.

SimulatedCounts dataclass

SimulatedCounts(
    tree: Tree, counts: ndarray, cell_totals: ndarray
)

UMI counts drawn on a known tree.

Attributes:

Name	Type	Description
tree	Tree	The generating tree.
counts	ndarray	(n_cells, n_genes) integer counts.
cell_totals	ndarray	The library size each cell was drawn with. Pass these, not the row sums: with a few hundred high-variance genes the row sums carry a per-cell compositional shift that costs the tree.