Quickstart¶
This tutorial walks you through a basic SCRIBE analysis. For a deeper understanding of the models and methods, see the Quick Overview and the Model Selection guide.
Basic Workflow¶
import scribe
import anndata as ad
# Load your scRNA-seq data
adata = ad.read_h5ad("data.h5ad")
# Run inference (default: NBVCP with variable capture)
results = scribe.fit(adata, n_steps=100_000)
# Inspect convergence
print(results.loss_history[-10:])
# Get posterior parameter distributions
distributions = results.get_distributions()
# Generate posterior predictive samples for model validation
ppc_samples = results.get_ppc_samples(n_samples=100)
# Visualize results
scribe.viz.plot_parameter_posteriors(results)
Choosing an inference method
Start with inference_method="svi" for fast iteration. Once you are
satisfied with the model choice, switch to inference_method="mcmc" for
publication-quality posterior distributions (requires GPU with double
precision support).
Next Steps¶
- Explore the Model Selection guide to find the right model for your data
- Learn about the Results class for downstream analysis
- See the
scribe.fit()Interface for full parameter documentation