Gene ontology(GO)-based autoencoder for embedding single-cell RNA-seq.

Generally the ontoencoder takes three input: X, y and topology (the gene ontology, or any directed acyclic graph you input)

How to use this directory

please refer to notebooks/TopoNet* for examples of supervised learning; notebooks/OntoEncoder* for unsupervised learning.

How to process my training data

Any single-cell RNA-seq can be log-normalized and saved as .h5ad by scanpy package. The processing step is recorded in notebooks/GSE71585-single cell.ipynb

Is there example data

processed data are stored in /cellar/users/hsher/ontoencoder/notebooks/tasic.h5ad (accessible to the Ideker lab)

How to prepare my ontology

the topology should be stored in the DCell format. See here for an example

This topology file can be converted to OntoEncoder/TopoNet-compatible format using ontoencoder.topology.topo_reader())

How to trim Gene Ontology(GO) into a smaller tree?

Please refer to OntoPrune [https://github.com/algaebrown/ontoPrune] for more information.

Can I train TopoNet/OntoEncoder on GPU?

I haven't implement that.

Dependency

environment.yml should be helpful. Refer to here how to install the same conda environment