[WIP] Kepler: VAE-based cell-type annotations

In this project, we predict cell-type annotations from an annotated dataset. We start with the PBMC 3K dataset from 10x Genomics (via Scanpy). This project is similar to our scRNA-seq Leukemia pipeline, but trains a VAE from scratch, rather than building on top of scVI.

Why a VAE + how a VAE works

VAE’s are useful for

Setting the project up

Training the model

Results

Turning it into a clinical grade model

Results

Conclusion