This repo is the final project for CS280: Deep generative model for auto-annotation in single-cell analysis. This project implemented a semi-supervised learning extended generative model for single-cell cell type annotation.

Compared with the SCANVI model, we adopted the idea of WAE. The code structure is mainly inspired by scvi

python annotationtest.py

data folder contaion two of our datasets used in the experiments: simulation_3.loom and high_data_loom.loom. The high_data_loom.loom is a dataset of mouse cells from different tissues which we merged by ourselves. The simulation_3.loom is a simulation dataset provided by scvi.

-annotationtest.py: the script to test annoation performance
-dgm4sca
|--dataset: scripys to load data
|--inference: scripts to classify cell type by posterior inference
|--models: scripts about generative model
-data: folder of data files in .loom format. (simulation_3.loom, high_data_loom.loom)

Torch version WAE

M1+M2 Model