BrainSTEM (brain Single-cell Two tiEr Mapping) is a biologically-driven multi-resolution mapping strategy to assess the transcriptional fidelity of midbrain cultures via scRNA-seq. In the first tier, query datasets are projected onto the fetal whole brain atlas to assess the brain region specificity of the various neural populations. Following this, midbrain-specific cells are further projected onto a more intricately annotated midbrain subatlas.
Here are some example codes to run BrainSTEM on the La Manno 2016 in vivo midbrain and our in-house midbrain differentiation dataset:
# Download ref objects
downloadWBref("path-to-save-the-whole-brain-atlas.rds")
downloadMBref("path-to-save-the-midbrain-subatlas.rds")
# Alternatively, the users can download required files by visiting the zenodo repository https://zenodo.org/records/17291276
# Read in ref and query seurat objects
seuRefWB <- readRDS("path-to-save-the-whole-brain-atlas.rds")
seuRefMB <- readRDS("path-to-save-the-midbrain-subatlas.rds")
# Run BrainSTEM on La Manno 2016
seuQueryManno <- readRDS("manno.rds")
seuQueryManno <- mapToWB(seuWBRef = seuRefWB, seuQuery = seuQueryManno)
seuQueryManno <- mapToMB(seuMBRef = seuRefMB, seuQuery = seuQueryManno, min.nCell = 0)
seuQueryMannoWBstats <- getWBstats(seuQueryManno, group.by = "group")
seuQueryMannoMBstats <- getMBstats(seuQueryManno, group.by = "group")
plotWBstats(seuQueryMannoWBstats)
plotMBstats(seuQueryMannoMBstats)
# Save plots
pdf(width = 6, height = 8.2, bg = "transparent", file = "plotMannoWBstats.pdf")
plotWBstats(seuQueryMannoWBstats)
dev.off()
pdf(width = 6, height = 12, bg = "transparent", file = "plotMannoMBstats.pdf")
plotMBstats(seuQueryMannoMBstats)
dev.off()
# Run BrainSTEM on in-house midbrain differentiation dataset
seuQueryToh <- readRDS("toh.inhouse.rds")
seuQueryToh <- mapToWB(seuWBRef = seuRefWB, seuQuery = seuQueryToh)
seuQueryToh <- mapToMB(seuMBRef = seuRefMB, seuQuery = seuQueryToh)
seuQueryTohWBstats <- getWBstats(seuQueryToh, group.by = "group")
seuQueryTohMBstats <- getMBstats(seuQueryToh, group.by = "group")
plotWBstat(seuQueryTohWBstats)
plotMBstat(seuQueryTohMBstats)To allow a convenient use of BrainSTEM, we provide downsampled atlases which contain 10% of the cells in the full whole brain/midbrain atlases. Internal benchmark shows that BrainSTEM using the downsampled atlases can recover 96.5%-100% celltype prediction in the first-tier projection and 88.7%-99.3% celltype prediction as compared to BrainSTEM using the full atlases. The downsampled atlases can be downloaded by:
downloadWBref("path-to-whole-brain-atlas.rds", mini = TRUE)
downloadMBref("path-to-midbrain-subatlas.rds", mini = TRUE)