script to make vizgen_cellbound toy dataset

inst/script/vizgen_cellbound.R

@@ -0,0 +1,396 @@
+#
+# load libs - less libs
+suppressPackageStartupMessages({
+  library(ggplot2)
+  #library(Seurat)
+  library(dplyr)
+  library(magrittr)
+  library(BiocParallel)
+  library(scuttle)
+  #library(SingleCellExperiment) 
+  #library(SpatialExperiment) 
+  library(SpatialFeatureExperiment)
+  library(Voyager)
+  library(terra)			
+  library(sf)
+})
+
+## ------------------------- ##
+## Brain cancer toy dataset 1
+## ------------------------- ##
+# my dir to large test dataset
+dir_use <- "./vizgen_test_repo/vizgen_cellbound/"
+dir_github <- "./vizgen_test_repo/vizgen_cellbound_github/"
+
+# load large SFE object
+sfe <-
+  readVizgen(data_dir = dir_use,
+             z = "all",
+             z_option = "3d", # this will return XYZ MULTIPOINT for rowGeometiries
+             sample_id = "vizgen_test",
+             min_area = 15,
+             image = c("DAPI", "PolyT", "Cellbound"),
+             flip = "geometry", # "image" & "none", "geometry"
+             max_flip = "50 MB",
+             filter_counts = TRUE, # keep cells with counts > 0 or not    
+             add_molecules = TRUE,
+             use_bboxes = FALSE,
+             #file_out = file.path(dir_use, "detected_transcripts.parquet"),
+             BPPARAM = BiocParallel::MulticoreParam(14, 
+                                                    tasks = 80L,
+                                                    force.GC = FALSE, 
+                                                    progressbar = TRUE)
+  )
+sfe
+# normalize raw counts
+sfe %<>% logNormCounts()
+sfe
+imgData(sfe)
+rowGeometry(sfe) %>% str # XYZ coords
+
+# check if flip works
+colGeometry(sfe, 1) %>% st_geometry() %>% st_bbox
+cellSeg(sfe) %>% st_geometry() %>% st_bbox
+txSpots(sfe) %>% st_geometry() %>% st_bbox
+
+# plot it
+# Segs
+options(repr.plot.height = 5, repr.plot.width = 10)
+pl1 <- 
+  plotSpatialFeature(sfe, features = "COL1A2", 
+                     size = 4,
+                     #colGeometryName = "centroids", 
+                     colGeometryName = "cellSeg",
+                     dark = TRUE,
+                     #scattermore = TRUE, # will plot only centroids!
+                     image_id = "Cellbound2_z3" # "DAPI_z3"
+  ) & Seurat::DarkTheme()
+pl1
+
+# Segs using bbox cropping
+bbox_use <- c(xmin = 6500, ymin = -1500, xmax = 6800, ymax = -1200)
+plotSpatialFeature(sfe, features = "COL1A2", bbox = bbox_use,
+                   size = 4, 
+                   #colGeometryName = "centroids",
+                   colGeometryName = "cellSeg",
+                   dark = TRUE,
+                   #scattermore = TRUE, # will plot only centroids!
+                   image_id = c("Cellbound2_z3"),
+) & Seurat::DarkTheme()
+#theme(axis.line = element_line(color = "white"), 
+#      axis.ticks = element_line(color = "white", size = 1, linetype = 1),
+#      axis.text = element_text(color = "white"))
+
+# subset obj given the bbox
+sfe_sub <- 
+  SpatialFeatureExperiment::crop(sfe, 
+                                 colGeometryName = "cellSeg",
+                                 sample_id = "vizgen_test",
+                                 y = bbox_use)
+# keep only background genes and some panel general markers
+sfe_sub <- sfe_sub[c(1:50, grep("Blank-", rownames(sfe_sub)))]
+sfe_sub
+
+# plot it - Segs
+options(repr.plot.height = 5, repr.plot.width = 10)
+pl1 <- 
+  plotSpatialFeature(sfe_sub, features = "COL1A2", 
+                     size = 4,
+                     #colGeometryName = "centroids", 
+                     colGeometryName = "cellSeg",
+                     dark = TRUE,
+                     #scattermore = TRUE, # will plot only centroids!
+                     image_id = "Cellbound2_z3" # "DAPI_z3"
+  ) & Seurat::DarkTheme()
+pl1
+
+# check if flip stays
+colGeometry(sfe_sub, 1) %>% st_geometry() %>% st_bbox
+cellSeg(sfe_sub) %>% st_geometry() %>% st_bbox
+txSpots(sfe_sub) %>% st_geometry() %>% st_bbox
+
+# export sfe_toy obj
+dir_local <- "./SpatialFeatureExperiment/seurat_v4/inst/extdata"
+saveRDS(sfe_sub, file = file.path(dir_local, "sfe_vizgen_toy.rds"))
+#sfe_sub <- readRDS(file = file.path(dir_local, "sfe_vizgen_toy.rds"))
+sfe_sub
+
+# OK -> get images ----
+imgData(sfe_sub)
+imgData(sfe_sub)[1,] # 1st image
+# get images in a list
+im_sub <- 
+  lapply(imgData(sfe_sub) %>% nrow() %>% seq(), 
+         function(i) {
+           im_sub <- 
+             Voyager:::.get_img_df(sfe_sub, 
+                                   sample_id = "vizgen_test", 
+                                   image_id = imgData(sfe_sub)$image_id[i], 
+                                   bbox = NULL # set bbox arg to NULL
+             )
+         })
+names(im_sub) <- imgData(sfe_sub)$image_id
+
+# get image - in a loop
+img <- 
+  lapply(seq(im_sub), function (i) {
+    im_sub[[i]]$data[[1]]@image %>% 
+      unwrap() # convert from PackedSpatRaster to SpatRaster
+  })
+plot(img[[1]])
+# convert to export as image
+img_out.list <-
+  lapply(seq(img), function(i) {
+    # using terra lib
+    img_out <- 
+      img[[i]] %>%
+      terra::aggregate(., fact = 4) %>%
+      #terra::flip() %>%
+      terra::as.array()
+    img_out[is.nan(img_out)] <- 0
+    return(img_out)
+  })
+# export images
+library(tiff)
+for (i in seq(img_out.list)) {
+  writeTIFF(img_out.list[[i]] / c(max(img_out.list[[i]]) * 3), # downsample image
+            paste0(dir_github, "images/mosaic_", names(im_sub)[i], ".tif"))
+}
+# load image and plot it
+img_test <- rast(file.path(dir_github, "images/mosaic_Cellbound1_z3.tif"))
+plot(img_test)
+
+# OK -> prepare manifest.json ----
+library(jsonlite)
+manifest <- read_json(file.path(dir_use, "images/manifest.json"), simplifyVector = TRUE)
+manifest %>% str
+mnfst <- manifest
+# keep only image names present in downsampled obj
+mnfst$mosaic_files %<>% 
+  filter(file_name %in% paste0("mosaic_", names(im_sub), ".tif"))
+mnfst$mosaic_pyramid_files <- NULL
+# extract relevant image dims
+img_out <- 
+  img[[1]] %>% 
+  terra::aggregate(., fact = 4) #%>% terra::flip()
+mnfst$mosaic_width_pixels <- ncol(img_out)
+mnfst$mosaic_height_pixels <- nrow(img_out)
+# make bbox with positive signes
+extent <- as.vector(ext(img_out))[c("xmin", "ymin", "xmax", "ymax")] 
+extent[c("ymin", "ymax")] <- -extent[c("ymax", "ymin")]	
+extent %>% unname()
+mnfst$bbox_microns <- extent %>% unname()
+mnfst$hor_num_tiles_box <- 1
+mnfst$vert_num_tiles_box <- 1
+mnfst %>% str
+write_json(mnfst, file.path(dir_github, "images/manifest.json"), 
+           pretty = TRUE,
+           auto_unbox = TRUE)
+
+# OK -> get count matrix ----
+mat_sub <- assay(sfe_sub, "counts")
+# read original count matrix
+mat <- data.table::fread(file.path(dir_use, "cell_by_gene.csv"), 
+                         colClasses = list(character = 1))
+mat %>% str
+# match them
+mat %<>%
+  # keep cells from mat_sub
+  dplyr::slice(match(mat_sub %>% colnames(), mat$cell)) %>%
+  # keep genes from mat_sub, and cell col
+  select(c(cell, 
+           match(mat_sub %>% rownames(), mat %>% colnames())))
+# export it
+data.table::fwrite(mat, file = file.path(dir_github, "cell_by_gene.csv"))
+
+# OK -> get metadata ----
+meta_orig <- data.table::fread(file.path(dir_use, "cell_metadata.csv"))
+meta_orig %>% str
+
+# get metadata df from sce, sfe or miloR obj
+callMeta <- function(object = NULL) {
+  return(colData(object)@listData %>% as.data.frame.list())
+}
+meta_sub <- 
+  callMeta(sfe_sub) %>% 
+  select(-contains("sample"))
+meta_sub %>% str
+
+# subset original metadata given cell ids
+cells_use <- 
+  match(meta_sub %>% rownames(), meta_orig$EntityID %>% as.character())
+# check if cell ids correspond
+identical(meta_orig %>%
+            dplyr::slice(cells_use) %>% 
+            pull(EntityID) %>% as.character, 
+          meta_sub %>% rownames())
+# all good!
+meta_sub <-
+  meta_orig %>%
+  dplyr::slice(cells_use)
+# export it
+data.table::fwrite(meta_sub, file = file.path(dir_github, "cell_metadata.csv"))
+
+# OK -> get cell segmentations ----
+parq_orig <- sfarrow::st_read_parquet(file.path(dir_use, "cell_boundaries.parquet"))
+parq_orig %>% str
+cellSeg(sfe_sub) %>% str
+
+# flip y coords
+mat_flip <- matrix(c(1,0,0,-1), ncol = 2)
+parq <- cellSeg(sfe_sub)
+st_geometry(parq) <- c(st_geometry(parq) * mat_flip)
+st_geometry(parq) %>% st_bbox
+
+# Not using that field anyway so can be random
+parq$ID <- sample(seq_len(nrow(parq)) - 1, nrow(parq))
+parq$EntityID <- 
+  cellSeg(sfe_sub) %>% 
+  rownames() %>% 
+  bit64::as.integer64.character(.)
+#parq$ZIndex <- 0
+#parq$Type <- "cell"
+#parq$ZLevel <- 1.5
+parq$ParentID <- parq$ParentType <- parq$Name <- NA
+parq$X__index_level_0__ <- parq$ID
+names(parq)[names(parq) == "geometry"] <- "Geometry"
+st_geometry(parq) <- "Geometry"
+parq <- parq[,names(parq_orig), drop = FALSE]
+parq %>% str
+# export file
+sfarrow::st_write_parquet(parq, 
+                          file.path(dir_github, "cell_boundaries.parquet")) %>% suppressWarnings()
+
+# OK -> get mols coords ----
+# XYZ coords
+rowGeometry(sfe_sub) %>% str
+rowGeometry(sfe_sub) %>% st_geometry() %>% str
+# flip mols coords
+txSpots(sfe_sub) %>% st_geometry() %>% st_bbox()
+mat_flip <- matrix(c(1,0,0,-1), ncol = 2)
+(st_geometry(txSpots(sfe_sub)) * mat_flip) %>% st_bbox
+mols <- txSpots(sfe_sub)
+st_geometry(mols) <- (st_geometry(mols) * mat_flip)
+# crop before convering mols
+mols <-
+  st_crop(mols, 
+          cellSeg(sfe_sub) %>%
+            st_geometry() %>%
+            st_bbox())
+mols %>% str
+
+# convert mols
+mols <-
+  bplapply(rownames(sfe_sub) %>% seq(), function(i) {
+    mols %>%
+      dplyr::slice(i) %>%
+      st_geometry() %>%
+      st_coordinates() %>%
+      as.data.frame() %>%
+      dplyr::transmute(global_x = X, 
+                       global_y = Y,
+                       #global_z = Z,
+                       gene = rownames(sfe_sub)[i])
+  }, BPPARAM = BiocParallel::MulticoreParam(12, 
+                                            tasks = 50L,
+                                            force.GC = FALSE, 
+                                            progressbar = TRUE)
+  ) %>% do.call(bind_rows, .)
+mols %>% str
+
+# load original molecule coords
+mols_orig <- data.table::fread(file.path(dir_use, "detected_transcripts.csv"))
+mols_orig %>% str
+
+# plot 1 molecule
+mols %>%
+  filter(gene == "COL1A2") %>%
+  ggplot(aes(global_x, global_y)) &
+  geom_hex(bins = 50)
+
+# filter orginal mols given cropped mols range
+bbox_mols <-
+  mols %>% 
+  select(contains("global")) %>%
+  apply(., 2, range)
+bbox_mols
+# filter
+mols_filt <-
+  filter(mols_orig %>% mutate(global_y = -global_y),
+         between(global_x, bbox_mols[1], bbox_mols[2]) & 
+           between(global_y, bbox_mols[3], bbox_mols[4]))
+mols_filt %>% str
+# join dfs
+mols_joined <-
+  dplyr::inner_join(mols, mols_filt)
+mols_joined %>% str
+# check range
+mols_joined %>% 
+  select(contains("global")) %>%
+  apply(., 2, range)
+
+# plot 1 molecule for all z-planes
+mols_joined %>%
+  filter(gene == "COL1A2") %>%
+  ggplot(aes(global_x, global_y, color = global_z)) &
+  geom_point(shape = 3) & 
+  # add previous plot
+  pl1
+# things seem to correspond, woo!
+
+# order vars
+mols_joined %<>%
+  # make y coord positive
+  mutate(global_y = -global_y) %>%
+  select(., names(mols_filt))
+mols_joined %>% str
+
+# export transcripts coords
+data.table::fwrite(mols_joined, file.path(dir_github, "detected_transcripts.csv"))
+
+# OK -> test loading the toy dataset ----
+dir_github <- "./vizgen_test_repo/vizgen_cellbound_github/"
+# load SFE object
+sfe <-
+  readVizgen(data_dir = dir_github,
+             z = "all",
+             z_option = "3d", # this will return XYZ MULTIPOINT for rowGeometiries
+             sample_id = "vizgen_toy",
+             min_area = 15,
+             image = c("DAPI", "PolyT", "Cellbound"),
+             flip = "geometry", # "image" & "none", "geometry"
+             max_flip = "50 MB",
+             filter_counts = TRUE, # keep cells with counts > 0 or not    
+             add_molecules = TRUE,
+             use_bboxes = FALSE,
+             #file_out = file.path(dir_use, "detected_transcripts.parquet"),
+             BPPARAM = BiocParallel::MulticoreParam(10, 
+                                                    tasks = 50L,
+                                                    force.GC = FALSE, 
+                                                    progressbar = TRUE)
+  )
+sfe
+# normalize raw counts
+sfe %<>% logNormCounts()
+sfe
+imgData(sfe)
+rowGeometry(sfe) %>% str # XYZ coords
+
+# toy obj
+colGeometry(sfe, 1) %>% st_geometry() %>% st_bbox
+cellSeg(sfe) %>% st_geometry() %>% st_bbox
+txSpots(sfe) %>% st_geometry() %>% st_bbox
+
+# plot it
+# Segs
+options(repr.plot.height = 5, repr.plot.width = 10)
+pl1 <- 
+  plotSpatialFeature(sfe, features = "COL1A2", 
+                     size = 4,
+                     #colGeometryName = "centroids", 
+                     colGeometryName = "cellSeg",
+                     dark = TRUE,
+                     #scattermore = TRUE, # will plot only centroids!
+                     image_id = "Cellbound2_z3", #"DAPI_z3",
+  ) & Seurat::DarkTheme()
+pl1