Cell Activities

Each cell is assigned a continuous activity score for every Splicing Program. These scores live in model.assign_post (or adata.obsm["X_PHI"]) and can be used for visualization, clustering, and downstream analysis exactly like a standard dimensionality reduction.

Accessing cell activities

import numpy as np

# (cells × K) — rows are cells, columns are splicing programs
phi = np.asarray(model.assign_post)

# Also written into adata:
phi = adata.obsm["X_PHI"]   # (cells × K)

UMAP / visualization

The X_PHI slot integrates directly with scanpy:

import scanpy as sc

adata.obsm["X_PHI"] = phi

sc.pp.neighbors(adata, use_rep="X_PHI")
sc.tl.umap(adata)
sc.pl.umap(adata, color=["cell_ontology_class", "tissue"])

Plotting activity per cell type

import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

df = pd.DataFrame(phi, columns=[f"SP_{k+1}" for k in range(phi.shape[1])])
df["cell_type"] = adata.obs["cell_ontology_class"].values

df_long = df.melt(id_vars="cell_type", var_name="SP", value_name="activity")
sns.boxplot(data=df_long, x="SP", y="activity", hue="cell_type")
plt.xticks(rotation=45)

Relationship to PSI loadings

A high activity score for SP k in a cell means that cell's splicing landscape is well explained by the PSI vector psi[k, :]. It does not mean the cell exclusively uses SP k — activities are soft assignments summing to approximately 1 across programs.