Developer guild

Note

To better understand the following guide, you may check out our publication first to learn about the general idea.

Below we describe main components of the framework, and how to extend the existing implementations.

Main components

A LTNN model is primarily composed of 3 components (all by Keras).

• ANN-A framework: to produce a pre-trainning model by existing scVelo latent time dataset and help researchers to determined the start and end node.

• ANN-B framework: to regress the pseudo time by the norm distribution of start and end node

• PAGA graph: to produce the pseudotime by PAGA trajectory

Current implementation for these components are all located in scltnn.models.sc. New extensions can be added to this module as well.