Overview

Overview#

Welcome to AnnDictionary#

Convenient parallel processing of anndata.

Why we built it#

We wanted to make it easier to analyze atlas-scale single-cell RNA sequencing data. When using scanpy and anndata, we found three friction points when scaling up analysis.

  1. We frequently wanted to do a parallel analysis separately on each stratum of an adata, and found ourselves copy-and-pasting a lot of code to do this.

    • Example: “Normalize each cell type separately, then look for differentially expressed genes in each, then plot.”

  2. Cell type and biological process annotation.

    • Example: “Label these cell types. Try 3 different strategies, and compare them all.”

  3. Managing labels within across datasets.

    • Example 1: “I downloaded data from 3 different groups, and they each used a different label for macrophages (i.e. macrophage, Macrophage., MΦ)”

    • Example 2: “I want to coarsen the cell type labels.”

So, we built a backend called AnnDictionary aimed at making these tasks easier.

Modules#

AnnDictionary is organized into the following modules:

  1. Core: This module contains the AdataDict class and looping constructs.

  2. LLM Management: This module handles LLM configuration and use, offering support for almost all mainstream LLM providers.

  3. Annotate: A suite of functions to annotate cells and genes with different strategies, including de novo (with LLMs) and by label transfer.

  4. Plot: Plotting functions (mostly ones that help annotation, some that help with gene list interpretation).

  5. Automated Label Management: For dealing with annoying category label problems.

  6. Utility Functions: Various utility and filtering functions.

  7. Wrappers: Pre-wrapped functions from packages that are designed to work on single adata.

The public API (i.e., how you, the end user, should use this package) is documented in the API section of the docs.

The source code for all functions (public and not) is available on GitHub.

Tutorials#

We offer a number of tutorial notebooks to allow you to familiarize yourself with using AnnDictionaries and how to adapt/scale up existing code. We also offer guides on using AnnDictionary to annotate cells.

Contribute#

The contribute section describes the architecture of this package in more detail, including guides on how to extend the LLM support and offer wrappers for other packages.

Cite#

If you use this package, please cite:

Benchmarking Cell Type Annotation by Large Language Models with AnnDictionary.
George Crowley, Tabula Sapiens Consortium, Stephen R. Quake.
bioRxiv 2024.10.10.617605
(https://doi.org/10.1101/2024.10.10.617605)