scArches (PyTorch) - single-cell architecture surgery

scArches is a package to integrate newly produced single-cell datasets into integrated references atlases. Our method can facilitate large collaborative projects with decentralised training and integration of multiple datasets by different groups. scArches is compatible with scanpy. and hosts efficient implementations of all conditional generative models for single-cell data.

What can you do with scArches?

• Integrate many single-cell datasets and share the trained model and the data (if possible).

• Download a pre-trained model for your atlas of interest, update it wih new datasets and share with your collaborators.

• Construct a customized reference by downloading a reference atlas, add a few pre-trained adaptors (datasets) and project your own data in to this customized reference atlas.

• Project and integrate query datasets on the top of a reference and use latent representation for downstream tasks, e.g.:diff testing, clustering.

What is an adaptor?

In scArches, each query datasets is added to the reference model by training a set of weights called adaptor. Each adaptor is a sharable object. This will enable users to download a reference model, customise that reference model with a set of adaptors (datasets) and finally add user data as a new adaptor and also share this adaptor with others.

Where to start?

To get a sense of how the model works please go through this tutorial. To find out how to construct and share or use pre-trained models example sections. Check this example to learn how to start with a raw data and pre-process data for the model.

It is always good to have a look at our training tips.