This repository is describing the construction of a scRNA-seq human T cell ProjecTILs reference. This reference aims at being comprehensive/exhaustive but also interpretable and robust.
-
To construct our model, training data were obtained from filtered data derived from Nick Borcherding collection.
-
CD8+ T cells were isolated using scGate.
-
Cycling cells and ISG+ cells were removed using UCell.
-
To improve integration, semi-supervised STACAS was run using pre-defined scGate classes as prior knowledge.
This ProjecTILs reference map is a ressource that enables rapid cluster annotation across many studies and patients, ultimately decreasing cell annotation time and allowing to draw parallels between studies, diseases and species.
To help integrating many datasets, scGate classes for the main CD8 T subsets were designed according to classical immunology markers. The graph below disaplys the classes used, with key marker genes for each cell type.
The full list of scGate markers used can be found here.
Integration was performed using ssSTACAS and scGate as prior annotation to help integration. This is especially critical for subsets which are hard to resolve, such as Progenitors Exhausted (TPEX).
For more projection examples, you can find them here: ProjecTILs_CaseStudies.