preprocesamiento dataset problema

ComunidadBioInfo · Aug 7, 2023 · 3c648e1 · 3c648e1
1 parent 1b60142
commit 3c648e1
Showing 1 changed file with 204 additions and 1 deletion.
diff --git a/02_sesion5.Rmd b/02_sesion5.Rmd
@@ -9,7 +9,7 @@ Erick Cuevas
 [
 
 ```{r,echo=FALSE}
-knitr::include_url("https://comunidadbioinfo.github.io/cdsb2023/brett_talk.pdf", height = "380px")
+knitr::include_url("https://erickcufe.github.io/anotacion_scRNA/anotacion_celulas.html", height = "380px")
 ```
 
 ](<https://comunidadbioinfo.github.io/cdsb2023/brett_talk.pdf>)
@@ -72,6 +72,7 @@ Dentro de este contexto, la anotación de clusters en scRNAseq es esencial. Una
 ## Paqueterías de R mas "famosas" para anotar
 
 - [SingleR](https://bioconductor.org/packages/release/bioc/vignettes/SingleR/inst/doc/SingleR.html)
+- [Seurat](https://satijalab.org/seurat/)
 - [scCATCH](https://github.com/ZJUFanLab/scCATCH)
 - [cellassign](https://github.com/Irrationone/cellassign)
 - [SCINA](https://github.com/jcao89757/SCINA)
@@ -90,3 +91,205 @@ Otros sitios interesantes para obtener conjuntos de datos de referencia:
 - [GLIASEQ](https://www.liddelowlab.com/gliaseq)
 
 
+## Preparación del dataset
+
+El dataset proviene tiene el ID en GEO [GSE84465](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE84465)
+
+```{r, eval=FALSE}
+library(cowplot)
+library(ggplot2)
+library(scater)
+library(scds)
+library(SingleCellExperiment)
+
+#Carga de datos y re formato
+#Carga de las cuentas crudas
+#Recuerda cambiar el PATH o directorio a la ubicación de los datos
+fastq_dirs <- list.dirs("../datos_atherosclerosis/tom_alsaigh_2022/", recursive = FALSE, full.names = TRUE)
+names(fastq_dirs) <- basename(fastq_dirs)
+sce <- DropletUtils::read10xCounts(fastq_dirs)
+
+
+#Renombrar row/colData nombre de columnas y SCE dimnames
+names(rowData(sce)) <- c("ENSEMBL", "SYMBOL")
+names(colData(sce)) <- c("sample_id", "barcode")
+sce$sample_id <- factor(basename(sce$sample_id))
+dimnames(sce) <- list(
+  with(rowData(sce), paste(SYMBOL, sep = ".")),
+  with(colData(sce), paste(barcode, sample_id, sep = ".")))
+
+
+# Agregamos la metadata
+
+md <- data.frame(
+  Sample_ID = c("Patient1_AC", "Patient1_PA",
+                "Patient2_AC", "Patient2_PA",
+                "Patient3_AC", "Patient3_PA"),
+  Age = c(82, 82, 87, 87, 65, 65),
+  AHA_Classification = c("Type_VII_Calcified", "Type_VII_Calcified",
+                         "Type_VII_Calcified", "Type_VII_Calcified",
+                         "Type_VII_Calcified", "Type_VII_Calcified"),
+  Region = c("Atherosclerotic_Core", "Proximal_Adjacent",
+             "Atherosclerotic_Core", "Proximal_Adjacent",
+             "Atherosclerotic_Core", "Proximal_Adjacent")
+)
+
+
+m <- match(sce$sample_id, md$Sample_ID)
+sce$region <- md$Region[m]
+sce$AHA_Classification <- md$AHA_Classification[m]
+sce$sample_id <- md$Sample_ID[m]
+sce$Age <- md$Age[m]
+
+```
+
+Hacemos ahora un pequeño proceso de limpieza y normalización
+
+```{r, eval=FALSE}
+#Remover genes no detectados
+sce <- sce[Matrix::rowSums(counts(sce) > 0) > 0, ]
+dim(sce)
+
+#Remover doublets
+#dividir SCE por muestra OJO esto podria ser por lotes
+cs_by_s <- split(colnames(sce), sce$sample_id)
+sce_by_s_2 <- lapply(cs_by_s, function(cs) sce[, cs])
+
+#correr 'scds'
+sce_by_s_2 <- lapply(sce_by_s_2, function(u)
+  scds::cxds_bcds_hybrid(scds::bcds(scds::cxds(u))))
+
+#Eliminando doublets
+sce_by_s <- lapply(sce_by_s_2, function(u) {
+  #Calcula numero de doublets (10x)
+  n_dbl <- ceiling(0.01 * ncol(u)^2 / 1e3)
+  #Elimina 'n_dbl' celulas con mayor scpre de doublet
+  o <- order(u$hybrid_score, decreasing = TRUE)
+  u[, -o[seq_len(n_dbl)]]
+})
+
+#Integrar nuevamente al objeto SCE
+sce <- do.call(cbind, sce_by_s)
+
+#Calcular metricas QC 
+(mito <- grep("MT-", rownames(sce), value = TRUE))
+# sce <- perCellQCMetrics(sce, subsets = list(Mito=mito))
+
+sce <- addPerCellQCMetrics(sce, subsets = list(Mito=mito))
+sce <- addPerFeatureQCMetrics(sce)
+
+```
+
+Filtración de células
+
+```{r, eval=FALSE}
+#Obtener outliers
+cols <- c("total", "detected", "subsets_Mito_percent")
+log <- c(TRUE, TRUE, FALSE)
+type <- c("both", "both", "higher")
+
+# Con esto decidimos que barcode desechar
+drop_cols <- paste0(cols, "_drop")
+for (i in seq_along(cols)){
+  colData(sce)[[drop_cols[i]]] <- isOutlier(sce[[cols[i]]],
+                                            nmads = 2.5, type = type[i],
+                                            log = log[i], batch = sce$sample_id)
+}
+
+# Muestra un resumen de los barcodes que se eliminaran
+sapply(drop_cols, function(i)
+  sapply(drop_cols, function(j)
+    sum(sce[[i]] & sce[[j]])))
+
+cd <- data.frame(colData(sce))
+ps <- lapply(seq_along(cols), function (i) {
+  p <- ggplot(cd, aes_string(x = cols[i], alpha = drop_cols[i])) +
+    geom_histogram(bins = 100, show.legend = FALSE) +
+    scale_alpha_manual(values = c("FALSE" = 1, "TRUE" = 0.4)) +
+    facet_wrap(~sample_id, ncol = 1, scales = "free") +
+    theme_classic() + theme(strip.background = element_blank())
+  if (log[i])
+    p <- p + scale_x_log10()
+  return(p)
+})
+plot_grid(plotlist = ps, ncol = 3)
+
+layout(matrix(1:2, nrow = 1))
+ol <- Matrix::rowSums(as.matrix(colData(sce)[drop_cols])) != 0
+x <- sce$total
+y <- sce$detected
+LSD::heatscatter(x, y, log="xy", main = "unfiltered",
+                 xlab = "Total counts", ylab = "Non-zero features")
+LSD::heatscatter(x[!ol], y[!ol], log="xy", main = "filtered",
+                 xlab = "Total counts", ylab = "Non-zero features")
+
+#Generar resumen de celulas a mantener
+ns <- table(sce$sample_id)
+ns_fil <- table(sce$sample_id[!ol])
+print(rbind(
+  unfiltered = ns, filtered = ns_fil,
+  "%" = ns_fil / ns * 100))
+
+#Eliminar celulas outlier
+sce <- sce[, !ol]
+dim(sce)
+
+#count > 1
+sce <- sce[Matrix::rowSums(counts(sce) > 1) >= 20, ]
+dim(sce)
+
+```
+
+Agrupamiento, toma en cuenta que esta parte podría demorar bastante.
+
+```{r, eval=FALSE}
+#Load packages
+library(cowplot)
+library(Seurat)
+library(SingleCellExperiment)
+library(ggplot2)
+
+#INTEGRATE
+#Crear SeuratObject
+so <- CreateSeuratObject(
+  counts = counts(sce),
+  meta.data = data.frame(colData(sce)),
+  project = "Alsaigh_10x_data")
+
+#Dividir por muestra
+cells_by_sample <- split(colnames(sce), sce$sample_id)
+so <- lapply(cells_by_sample, function(i)
+  subset(so, cells = i))
+
+#Normalizar, encontrar genes variables, escalar
+so <- lapply(so, NormalizeData, verbose = FALSE)
+so <- lapply(so, FindVariableFeatures, nfeatures = 2e3,
+             selection.method = "vst", do.plot = FALSE, verbose = FALSE)
+so <- lapply(so, ScaleData, verbose = FALSE)
+
+#Encontrar anclas
+# Estas anclas nos ayudaran despues para integrar todo con la funcion IntegrateData.
+as <- FindIntegrationAnchors(so, verbose = TRUE)
+so <- IntegrateData(anchorset = as, dims = seq_len(30), verbose = TRUE)
+
+#Escalar datos integrados
+DefaultAssay(so) <- "integrated"
+so <- ScaleData(so, display.progress = FALSE)
+
+#Reducción de dimension
+so <- RunPCA(so, npcs = 100, verbose = FALSE)
+
+#Cambiar numbero de PCs usedos
+so <- RunTSNE(so, reduction = "pca", dims = seq_len(20),
+              seed.use = 1, do.fast = TRUE, verbose = FALSE)
+so <- RunUMAP(so, reduction = "pca", dims = seq_len(20),
+              seed.use = 1, verbose = FALSE)
+
+#CLUSTERING
+so <- FindNeighbors(so, reduction = "pca", dims = seq_len(20), verbose = FALSE)
+for (res in c(0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1)){
+  so <- FindClusters(so, resolution = res, random.seed = 1, verbose = FALSE)
+}
+
+```
+