An autoencoder for single cell data.
- Free software: MIT
- Documentation: https://sc-autoencoder.readthedocs.io.
This project uses an autoencoder model to learn latent features from single-cell RNA-seq (scRNA-seq) data. Autoencoder models and similar architectures are frequently used for scRNA-seq data. For instance, Eraslan et al. used an autoencoder for denoising of single cell data. In another study, Lotfollahi et al use a variational autoencoder to predict perturbation responses. Here, we have implemented a very simple autoencoder to demonstrate how non-deterministic operations can lead to significant differences in latent space embeddings which affect downstream analysis and hinder reproducibility.
The model used in this project follows a standard encoder-encoding-decoder autoencoder architecture. We use layer sizes of 256, 128 and 64 for the encoder and decoder (in reverse) layers, and a encoding size of 32.
This package was created with mlf-core using Cookiecutter.