juzi 0.0.7

juzi

:tangerine: juzi

Various methods for analyzing cell states and types in single-cell sequencing data (experimental).

• cell states (cs)
• marker genes (mg)

Installation

pip install juzi

cell states (cs)

Identifying intra-sample programs.

from juzi.cs.nmf import gaussian_nmf, poisson_nmf, fixed_gaussian_nmf, fixed_poisson_nmf

# Vanilla NMF on normalized and log1p counts
W, H, losses = gaussian_nmf(
    data,
    n_factors=8,
    max_iter=100,
    lambda_H=1e-2,
    init="random",
    eps=1e-7
) 

# Poisson NMF on counts
W, H, losses = poisson_nmf(
    data,
    n_factors=8,
    max_iter=100,
    lambda_H=1e-2,
    init="nndsvd",
    eps=1e-7
)

# Vanilla NMF with fixed H
W, losses = fixed_gaussian_nmf(
    data,
    fixed_H=H,
    max_iter=100,
    init="random",
    eps=1e-7,
    silent=False
)

# Poisson NMF with fixed H
W, losses = fixed_poisson_nmf(
    data,
    fixed_H=H,
    max_iter=100,
    init="random",
    eps=1e-7,
    silent=False
)

Identifying consensus (intra-sample) and shared (inter-sample) programs.

from juzi.cs.nmf import factor_consensus, factor_similarity

# Compute a set of clustered consensus factors between NMF runs
HC, HS, labels, correlation = factor_consensus(
    [H1, H2, H3, ...],
    n_clusters=10,
    eps=1e-8,
    method="agglomerative",
    metric="euclidean",
    linkage="ward",
)

# Compute similarity matrix between factors computed across different samples
S, K, ids = factor_similarity(
    [H1, H2, H3, ...],
    distance="cosine",
    top_k=500,
    drop_zeros=True,
    intra_sample=False,
    eps=1e-8
)

Some additional tools for clustering and filtering factors.

from juzi.cs.cluster import eigengap_heuristic, spectral_clustering

# Estimate an optimal number of clusters from a similarity matrix
results = eigengap_heuristic(
    S,
    min_clusters=2,
    max_clusters=10,
    normalize=True,
    eps=1e-8
)

results.plot(show=True)

# Cluster a similarity matrix given an input number of clusters
assignments = spectral_clustering(
    S,
    n_clusters=results.k,
    normalize=True,
    seed=123456,
    eps=1e-8
)

marker genes (mg)

Various marker genes for cell types, subtypes, and pathways.

from juzi.mg import available_sets

# Check available marker gene sets
print(available_sets())

# Load breast cancer gene sets (e.g. PAM50)
from juzi.mg import CancerBreast
markers = CancerBreast()

# Load cancer pathway gene sets (e.g. HIPPO)
from juzi.mg import CancerPathways
markers = CancerPathways()

# Load cell cycle gene sets (e.g. G1S)
from juzi.mg import CellCycle
markers = CellCycle()

# List available sets in a given marker class
markers.info()

# Get all genes from all sets in a flattened list
markers.all()