Merge pull request #80 from estellad/main

add reader for Visium HD cell segmented output

Author: lambdamoses

R/read.R

@@ -364,6 +364,151 @@ readVisiumHD <- function(data_dir, bin_size = c(2L, 8L, 16L),
     sfes
 }
 
+#' Flip the Y-axis of cell or nucleus segmentations to align with H&E image. 
+#'
+#' @param sf a sf object read from a `.geojson` file.
+#' @param type "POINT" for cell centroid, or "POLYGON" for cell segmentation 
+#' mask. Default is "POINT".
+#' @param img_height total length along the Y axis of the image. Obtained by 
+#' reading in `hires` or `lowre`s `.png` under `/spatial` folder with 
+#' `magick::image_read()`.
+#' @param scalef scaling factor from `/spatial/scalefactors_json.json` file
+#'
+#' @returns a sf object with Y-axis of the points or polygons flipped.
+#'
+#' @examples
+#' geo_data_flipped <- flip_sf_Y(sf = geo_data, type = "POLYGON", 
+#'                               img_height = 3886, scalef = 0.079)
+flip_sf_Y <- function(sf, type = "POINT", img_height, scalef){
+  st_geometry(sf) <- st_sfc( 
+    lapply(st_geometry(sf), function(geom) {
+      coords <- st_coordinates(geom)
+      coords[, 2] <- img_height/scalef - coords[, 2] 
+      if(type == "POINT"){
+        st_point(coords)
+      }else{ # type == "POLYGON"
+        st_polygon(list(matrix(coords[, 1:2], ncol = 2)))
+      }
+    }),
+    crs = st_crs(sf)
+  )
+  
+  return(sf)
+}
+
+
+#' Reader for Visium HD cell segmented output.
+#'
+#' @param td path to unzipped Visium HD data download with `/segmented_outputs`
+#' @param res "hires" or "lowres" for what .png to read in. 
+#'
+#' @returns a `SpatialFeatureExperiment` object.
+#'
+#' @examples
+#' vhdsfe <- readVisiumHDCellSeg(td = "~/Desktop", res = "hires")
+readVisiumHDCellSeg <- function(td, res){
+  if(res == "hires"){
+    png_name = "tissue_hires_image.png"
+    scalef_name = "tissue_hires_scalef"
+  }else{ # res = "lowres"
+    png_name = "tissue_lowres_image.png"
+    scalef_name = "tissue_lowres_scalef"
+  }
+  
+  ## Coord ---------------------------------------------------------------
+  # Read the GeoJSON file of cell segmentation
+  geo_data <- st_read(file.path(td, "segmented_outputs", 
+                                "cell_segmentations.geojson"))
+  scalef <- rjson::fromJSON(file = file.path(td, "segmented_outputs", "spatial", 
+                                             "scalefactors_json.json"))
+  image <- image_read(file.path(td, "segmented_outputs", 
+                                "spatial", png_name))
+  info <- image_info(image)
+  
+  # Get centroid
+  st_crs(geo_data) <- NA
+  rownames(geo_data) <- geo_data$cell_id
+  centroids <- st_centroid(geo_data)
+  centroids$cell_id <- as.character(centroids$cell_id)
+  
+  
+  ## Countmat ------------------------------------------------------------
+  countmat_file <- file.path(td, "segmented_outputs", 
+                             "filtered_feature_cell_matrix.h5")
+  vhdcellsce <- DropletUtils::read10xCounts(countmat_file, col.names = TRUE)
+  
+  colnames(vhdcellsce) <- sub("cellid_0*([0-9]+)-.*", "\\1", vhdcellsce$Barcode)
+  
+  rownames(vhdcellsce) <- rowData(vhdcellsce)$Symbol
+  rownames(vhdcellsce) <- make.unique(rownames(vhdcellsce))
+  
+  
+  ## Matching coordinates with count matrix ------------------------------
+  # Subset to common cells
+  common_cells <- intersect(centroids$cell_id, colnames(vhdcellsce))
+  
+  geo_data <- geo_data[geo_data$cell_id %in% common_cells, ]
+  centroids <- centroids[centroids$cell_id %in% common_cells, ]
+  vhdcellsce <- vhdcellsce[, colnames(vhdcellsce) %in% common_cells]
+  
+  # # Sanity check on ordering of cell ids
+  # all(as.character(centroids$cell_id) == colnames(vhdcellsce))
+  
+  # Extract coordinates
+  coords <- st_coordinates(centroids)
+  colnames(coords) <- c("pxl_col_in_fullres", "pxl_row_in_fullres")
+  
+  
+  ## Construct SPE -------------------------------------------------------
+  vhd <- SpatialExperiment(
+    assays = list(counts = as(counts(vhdcellsce), "dgCMatrix")),
+    rowData = rowData(vhdcellsce),
+    colData = cbind(colData(vhdcellsce), coords),
+    spatialCoordsNames = colnames(coords),
+    scaleFactors = scalef$tissue_hires_scalef, 
+    imageSources = file.path(td, "segmented_outputs", "spatial", png_name), 
+    loadImage = TRUE
+  )
+  imgData(vhd)$image_id <- res
+  
+  # vhd
+  
+  ## Coerce to SFE -------------------------------------------------------
+  vhdsfe <- toSpatialFeatureExperiment(vhd)
+  # vhdsfe
+  
+  ## Add polygons to colGeometries
+  colGeometries(vhdsfe)$cellSeg <- geo_data
+  
+  
+  # Flip Y in colGeometries -------------------------------------------------
+  # Flip Centroid Y
+  centroids <- colGeometries(vhdsfe)$centroids
+  colGeometries(vhdsfe)$updatecentroids <- 
+    flip_sf_Y(centroids, type = "POINT", img_height = info$height, 
+              scalef = imgData(vhdsfe)$scaleFactor) 
+  
+  # Flip Cellseg Y
+  cellSeg <- colGeometries(vhdsfe)$cellSeg
+  colGeometries(vhdsfe)$updatecellSeg <- 
+    flip_sf_Y(cellSeg, type = "POLYGON", img_height = info$height, 
+              scalef = imgData(vhdsfe)$scaleFactor) 
+  
+  # Flip Centroid Y in `spatialCoords()`
+  spatialCoords(vhdsfe)[, "pxl_row_in_fullres"] <- 
+    info$height/imgData(vhdsfe)$scaleFactor - 
+    spatialCoords(vhdsfe)[, "pxl_row_in_fullres"]
+  
+  # Add high res coords to `colData()`
+  vhdsfe[[paste0("pxl_col_in_", res)]] <- 
+    spatialCoords(vhdsfe)[, "pxl_col_in_fullres"] * imgData(vhdsfe)$scaleFactor
+  vhdsfe[[paste0("pxl_row_in_", res)]] <- 
+    spatialCoords(vhdsfe)[, "pxl_row_in_fullres"] * imgData(vhdsfe)$scaleFactor
+  
+  return(vhdsfe)
+}
+
+
 #' @importFrom sf st_nearest_feature st_distance
 #' @importFrom stats median
 .pixel2micron <- function(df) {