Welcome to SCRIBE’s documentation!
Warning
Project Status and Usage Restrictions
This project is currently in a pre-release state and is made available for viewing and evaluation purposes only.
At this time:
You may view and evaluate the code and documentation
You may not use, copy, modify, or distribute any part of this software
You may not incorporate this code into other projects
The software will be released under the MIT License following the publication of the associated pre-print. Until then, all rights are reserved.
SCRIBE (Single-Cell RNA-Seq Inference using Bayesian Estimation) is a Python package for analyzing single-cell RNA sequencing (scRNA-seq) data using variational inference based on Numpyro—a Jax-based probabilistic programming library with GPU acceleration. It provides a collection of probabilistic models and inference tools specifically designed for scRNA-seq count data.
Features
Multiple probabilistic models for scRNA-seq data analysis
Support for both full-batch and mini-batch inference for large-scale data
Integration with AnnData objects
Comprehensive visualization tools for posterior analysis
GPU acceleration support
Available Models
SCRIBE includes several probabilistic models for scRNA-seq data, all documented in detail in SCRIBE Models for Single-Cell RNA Sequencing:
Core Model: Negative Binomial-Dirichlet Multinomial (NBDM)
Models both count magnitudes and proportions
Accounts for overdispersion in count data
Forms the foundation of SCRIBE’s modeling approach
Zero-Inflated Negative Binomial (ZINB)
Handles excess zeros in scRNA-seq data
Models technical and biological dropouts
Includes gene-specific dropout rates
Negative Binomial with Variable Capture Probability (NBVCP)
Accounts for cell-specific mRNA capture efficiency
Models technical variation in library preparation
Suitable for datasets with varying sequencing depths per cell
Zero-Inflated Negative Binomial with Variable Capture Probability (ZINBVCP)
Combines zero-inflation and variable capture probability
Most comprehensive model for technical variation
Handles both dropouts and capture efficiency
All these models can be extended to mixture variants as documented in Mixture Models to account for heterogeneous cell populations.