Decima

Introduction

Decima is a Python library to train sequence models on single-cell RNA-seq data.

Weights

Weights of the trained Decima models (4 replicates) are now available at https://zenodo.org/records/15092691. See the tutorial for how to load and use these.

Preprint

Please cite https://www.biorxiv.org/content/10.1101/2024.10.09.617507v3. Also see https://github.com/Genentech/decima-applications for all the code used to train and apply models in this preprint.

Requirements

Decima has been tested on Ubuntu 24.04.3 and MacOS 15.6.1 using Python 3.9-3.12.

Installation

Install the package from PyPI,

pip install decima

Or if you want to be on the cutting edge,

pip install git+https://github.com/genentech/decima.git@main

Typical installation time including all dependencies is under 10 minutes.

Tutorials

See the tutorials for instructions, including how to train your own Decima model with an example dataset.

Note

This project has been set up using BiocSetup and PyScaffold.