make.projection: Project a query scRNA-seq dataset onto a reference atlas

In carmonalab/ProjecTILs: Reference-based analysis of scRNA-seq data

Project a query scRNA-seq dataset onto a reference atlas

Description

This function allows projecting ("query") single-cell RNA-seq datasets onto a reference map (i.e. a curated and annotated scRNA-seq dataset). To project multiple datasets, submit a list of Seurat objects with the query parameter. The projection consists of 3 steps:

• pre-processing: optional steps which might include pre-filtering of cells by markers using 'scGate', data normalization, and ortholog conversion.

• batch-effect correction: uses built-in STACAS algorithm to detect and correct for batch effects (this step assumes that at least a fraction of the cells in the query are in the same state than cells in the reference)

• embedding of corrected query data in the reduced-dimensionality spaces (PCA and UMAP) of the reference map.

Usage

make.projection(
  query,
  ref = NULL,
  filter.cells = TRUE,
  query.assay = NULL,
  direct.projection = FALSE,
  STACAS.anchor.coverage = 0.7,
  STACAS.correction.scale = 100,
  STACAS.k.anchor = 5,
  STACAS.k.weight = "max",
  skip.normalize = FALSE,
  fast.umap.predict = FALSE,
  ortholog_table = NULL,
  scGate_model = NULL,
  ncores = 1,
  progressbar = TRUE
)

Arguments

query	Query data, either as single Seurat object or as a list of Seurat object
ref	Reference Atlas - if NULL, downloads the default TIL reference atlas
filter.cells	Pre-filter cells using 'scGate'. Only set to FALSE if the dataset has been previously subset to cell types represented in the reference.
query.assay	Which assay slot to use for the query (defaults to DefaultAssay(query))
direct.projection	If true, apply PCA transformation directly without alignment
STACAS.anchor.coverage	Focus on few robust anchors (low STACAS.anchor.coverage) or on a large amount of anchors (high STACAS.anchor.coverage). Must be number between 0 and 1.
STACAS.correction.scale	Slope of sigmoid function used to determine strength of batch effect correction.
STACAS.k.anchor	Integer. For alignment, how many neighbors (k) to use when picking anchors.
STACAS.k.weight	Number of neighbors to consider when weighting anchors. Default is "max", which disables local anchor weighting.
skip.normalize	By default, log-normalize the count data. If you have already normalized your data, you can skip normalization.
fast.umap.predict	Fast approximation for UMAP projection. Uses coordinates of nearest neighbors in PCA space to assign UMAP coordinates (credits to Changsheng Li for the implementation)
ortholog_table	Dataframe for conversion between ortholog genes (by default package object Hs2Mm.convert.table)
scGate_model	scGate model used to filter target cell type from query data (if NULL use the model stored in ref@misc$scGate)
ncores	Number of cores for parallel execution (requires BiocParallel)
progressbar	Whether to show a progress bar for projection process or not (requires BiocParallel)

Details

See load.reference.map to load or download a reference atlas. See also ProjecTILs.classifier to use ProjecTILs as a cell type classifier.

Value

An augmented Seurat object with projected UMAP coordinates on the reference map

Examples

data(query_example_seurat)
ref <- load.reference.map()
make.projection(query_example_seurat, ref=ref)

carmonalab/ProjecTILs documentation built on April 13, 2025, 10:04 p.m.