carmonalab/ProjecTILs: ProjecTILs-3.0

New in this version:

• new function make.reference for easy conversion of custom Seurat objects into reference maps for ProjecTILs. See custom map tutorial for an example (not T cells this time :))
• new function Run.ProjecTILs as a wrapper for projection (make.projection) and cell type prediction (cell.states.predict).
• new function ProjecTILs.classifier to transfer labels to query without altering the embeddings of the query dataset
• new split.by option in Run.ProjecTILs and ProjecTILs.classifier to split datasets by a grouping variable (normally the sample or the sequencing batch) before projection; results are then re-merged after projection into a single object.
• new function recalculate.embeddings to recompute PCA and/or UMAP after projection, accounting for both reference and query cells. This may be used as a way to update a reference when a query dataset contains a novel cell type / cell state.
• updates to the projection algorithm to use batch-effect correction as in STACAS-2.0. This includes anchor weighting based on rPCA distance, and STACAS.k.weight="max" by default, which disables local re-weighting of integration anchors.
• support for uwot for UMAP embeddings in new references generated with make.reference; the projection algorithm automatically detects which algorithm was used for reference building and applies the same method for query data embedding. NOTE: there are known issues with saving uwot models to .rds files (see e.g. this issue) - the default method in ProjecTILs remains umap until this issue is solved.
• allow plotting signature scores on radar plots. Set using meta4radar parameter in plot.states.radar function.
• keep specified order of genes in plot.states.radar (previously was automatically reordered alphabetically)
• exposing function merge.Seurat.embeddings (previously available as internal function): a utility function that merges two Seurat objects as well as their low dimensional embeddings
• allow specifying a custom list of genes to find.discriminant.genes, e.g. to exclude pseudo-genes or ribosomal genes from DEG analysis