Skip to content

User Guide

This section covers the practical workflow of using SCRIBE---from selecting a model through downstream analysis.

  • The scribe.fit() Interface

    The single entry point for all SCRIBE inference, with every parameter group explained, code examples, and links to deeper pages

    The scribe.fit() Interface

  • Model Selection

    Choosing the right model: NB base, zero inflation, variable capture, BNB overdispersion, mixture components, and parameterizations

    Model Selection

  • Parameter Reference

    Color-coded cheatsheet mapping every internal parameter name to its symbol, equation context, biological meaning, and parameterization

    Parameter Reference

  • Variational Guide Families

    Mean-field, low-rank, joint low-rank, normalizing flows, amortized, and VAE latent guides: what they capture and when to use each

    Variational Guide Families

  • Inference Methods

    Choosing between SVI, MCMC, and VAE, key parameters, early stopping, and the SVI-to-MCMC warm-start workflow

    Inference Methods

  • Results Class

    Understanding and using ScribeResults for posterior analysis, sampling, denoising, and normalization

    Results Class

  • Differential Expression

    Bayesian DE with three methods, error control via lfsr and PEFP, biological-level metrics, gene masking, and pathway analysis

    Differential Expression

  • Model Comparison

    WAIC, PSIS-LOO, stacking weights, per-gene goodness-of-fit diagnostics, and integration with the DE pipeline

    Model Comparison

  • scribe-infer CLI

    Reproducible, config-driven inference via Hydra with SLURM integration and automatic covariate-split orchestration

    scribe-infer CLI

  • scribe-visualize CLI

    Post-inference diagnostic plots: loss curves, ECDF, PPC grids, UMAP overlays, heatmaps, and more --- with recursive and SLURM support

    scribe-visualize CLI