--- title: "Farris et al RNA-seq methods" output: html_document: code_folding: hide toc: true toc_depth: 4 toc_float: collapsed: true theme: cerulean highlight: pygments df_print: kable word_document: fig_width: 6 fig_height: 6 fig_caption: true always_allow_html: yes --- ```{r setup, include=FALSE} # css: liao_style.css ## To render this page: ## In the farrisSeq folder run: ## rmarkdown::render("farrisSeq.Rmd", "html_document", knit_root_dir=getwd(), output_dir=getwd()); knitr::opts_chunk$set(cache=TRUE, autodep=TRUE, cache.extra=list( packageVersion("farrisdata"), packageVersion("jamba"), packageVersion("splicejam")) ); options("kable_styling_position"="left", "kable_styling_full_width"=FALSE, "kable_styling_bootstrap_options"=c("striped", "hover", "condensed", "responsive") ); ``` This document describes and demonstrates analysis methods used to analyze mouse hippocampal subregion- and compartment-specific transcriptome RNA-seq data [(Farris et al, 2019)](https://www.cell.com/cell-reports/fulltext/S2211-1247(19)31155-6). Raw sequence read data is provided through GEO GSE116343, with RNA-seq data available at [GEO accession GSE116342](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE116342). Data used for analysis in R is provided through an R package `farrisdata`, installed in R with `devtools::install_github("jmw86069/farrisdata")`. ## Reproducing this document This document can be reproduced using the R code displayed in this document. There are three basic methods to reproduce these steps. #### Method One Copy and paste the commands into an active R session. Note that the workflow will create temporary files in the current directory. You can check the active directory with `getwd()` in R. #### Method Two Download the Rmarkdown file [farrisSeq.Rmd](https://raw.githubusercontent.com/jmw86069/jampack/master/docs/farrisSeq.Rmd) (linked) from the Github repository. * Move this file to a convenient folder. Cache files and images will be stored in that folder as a result of the analysis steps below. * Open R in the folder where the file was saved, then run this command in R: > rmarkdown::render("farrisSeq.Rmd", "html_document", knit_root_dir=getwd(), output_dir=getwd()) Once the process is complete, there will be a new file `farrisSeq.html` that contains the full summary including figures. This method has the benefit of loading all the R objects into the active R environment, so the data can be examined in more detail. #### Method Three Download the Rmarkdown file [farrisSeq.Rmd](https://github.com/jmw86069/jampack/blob/master/docs/farrisSeq.html) (linked) from the Github repository. * Move this file to a convenient folder. Cache files and images will be stored in that folder as a result * Open the `farrisSeq.Rmd` file in Rstudio, and click `"Knit"` to render the document into an HTML page. Once this process is complete, the output file `farrisSeq.html` will be opened automatically in the Rstudio viewer. This process has the slight disadvantage that the R objects are not loaded into the R environment. ## R environment setup The initial steps involve loading required R packages into memory. ```{r, r_initialization, cache=FALSE} suppressPackageStartupMessages(library(knitr)); suppressPackageStartupMessages(library(kableExtra)); suppressPackageStartupMessages(library(dplyr)); suppressPackageStartupMessages(library(tidyselect)); suppressPackageStartupMessages(library(plotly)); suppressPackageStartupMessages(library(made4)); suppressPackageStartupMessages(library(matrixStats)); suppressPackageStartupMessages(library(ComplexHeatmap)); suppressPackageStartupMessages(library(SummarizedExperiment)); suppressPackageStartupMessages(library(ggplot2)); suppressPackageStartupMessages(library(ggrepel)); suppressPackageStartupMessages(library(jamba)); suppressPackageStartupMessages(library(jamma)); suppressPackageStartupMessages(library(colorjam)); suppressPackageStartupMessages(library(splicejam)); suppressPackageStartupMessages(library(farrisdata)); suppressPackageStartupMessages(library(png)); suppressPackageStartupMessages(library(Cairo)); ``` ## Description of Input Data The input data for this analysis workflow is gene and transcript level read count data derived from Salmon quantitation files, which were already imported and assembled into relevant data matrices and saved as R objects `farrisGeneSE` and `farrisTXSE`, as described in more depth below. ### Description of sample groups The cell body (CB) and dendrite (DE) compartments from each hippocampal subregion cell type (CA1, CA2, CA3, DG) from three adult male mice (animal IDs 492, 496, 502) were laser capture microdissected and total RNA was extracted for RNAseq. Libraries were made using the Nugen Ovation Universal RNA-Seq system, which does not use isothermal amplification and produces strand-specific libraries depleted of rRNA. For consistency, a common set of colors are defined to represent the different sample groups, obtained from `farrisdata::colorSub`. ### Gene expression data Salmon quantitation files were imported using the R Bioconductor package `tximport`, summarized to the gene level, and stored in an R object `farrisGeneSE`, which is available in the R data package `"farrisdata"`, installed using `devtools::install_github("jmw86069/farrisdata")`. ```{r load_gene_data} ## Load the gene data farrisGeneSE; ``` #### farrisGeneSE SummarizedExperiment format `farrisGeneSE` is a `SummarizedExperiment` object with this content: * **assays** list containing the numeric matrix with gene rows, sample columns, with log2 median-normalized counts. Each row is named using a unique gene symbol derived from the Gencode GTF file. To access the normalized Salmon pseudocounts: ```{r, eval=FALSE} assays(farrisGeneSE)[["counts"]] ``` * **assays** also contains the raw Salmon pseudocounts accessible, using this command: `assays(farrisGeneSE)[["raw_counts"]]` * **rowData** data.frame with gene rows, using the gene symbol derived from the Gencode GTF file. ```{r, gene_table, dependson="load_gene_data"} as.data.frame(head(rowData(farrisGeneSE), 10)) %>% dplyr::mutate() %>% dplyr::select(probeID, geneSymbol, tidyselect::ends_with("_type"), tidyselect::contains("has")) %>% kable(escape=FALSE) %>% kable_styling() ``` * **colData** annotated data.frame with sample rows, describing the mouse hippocampal **Region**, **CellType**, **Sample**, and **groupName**. ```{r, sample_table, echo=FALSE, dependson=c("load_gene_data","color_sub")} colorSub <- farrisdata::colorSub; df2 <- kable_coloring(colData(farrisGeneSE), colorSub=colorSub, row_color_by="groupName", returnType="kable") %>% collapse_rows(columns=c(5), valign="middle") %>% row_spec(0, background="#DDDDDD") df2; ``` * **metadata** list describing the experimental design and contrasts used for statistical comparisons. * **design** numeric matrix describing the sample design, suitable for use directly by methods from the R Bioconductor `limma` package. * **contrasts** numeric matrix describing contrasts, suitable for use directly by `limma`. * **samples** vector of samples used. * **genes** vector of genes used. ### Transcript expression data Salmon quantitation files were imported using `tximport` as above, maintaining individual isoform abundances, then stored in an R object `farrisTxSE`, which is available in the R data package `farrisdata`, installed using `devtools::install_github("jmw86069/farrisdata")`. ```{r load_tx_data} ## Load the gene data farrisTxSE; ``` #### farrisTxSE format The `farrisTxSE` object is a named list with content similar to that described above for gene data: * **assays** list containing the numeric matrix with gene rows, sample columns, with log2 median-normalized counts. Each row is named using a unique gene symbol derived from the Gencode GTF file. * **rowData** data.frame with gene rows, using the gene symbol derived from the Gencode GTF file. ```{r, tx_table, dependson=c("load_tx_data","color_sub")} tx_df <- subset(as.data.frame(rowData(farrisTxSE)), geneSymbol %in% c("Gria1","Shank2","Ntrk2")) %>% dplyr::select(geneSymbol, transcript_id, tidyselect::ends_with("_type"), tidyselect::contains("has"), tidyselect::contains("tpm")); tx_df <- mixedSortDF(tx_df, byCols=match(c("geneSymbol","TxDetectedByTPM"), colnames(tx_df))*c(1,-1)); colorSubGene <- c(colorSub, group2colors(unique(tx_df$geneSymbol), cRange=c(20,30), lRange=c(80,90))); tx_df2 <- kable_coloring(tx_df, colorSub=colorSubGene, row_color_by="geneSymbol", verbose=FALSE, returnType="kable") %>% collapse_rows(columns=2, valign="middle") %>% row_spec(0, background="#DDDDDD") tx_df2; ``` * **colData** annotated data.frame with sample rows, describing the mouse hippocampal **Region**, **CellType**, **Sample**, and **groupName**. This data is identical to the sample table shown for `farrisGeneSE` above. * **metadata** list describing the experimental design and contrasts used for statistical comparisons. * **design** numeric matrix describing the sample design, suitable for use directly by methods from the R Bioconductor `limma` package. * **contrasts** numeric matrix describing contrasts, suitable for use directly by `limma`. * **samples** vector of samples used. * **genes** vector of genes used. ## Related analysis parameters The following steps and parameters were used to create the Salmon median normalized count data used in differential gene and isoform expression analyses and/or STAR junction count data used in sashimi plots. The STAR alignment was also used to create sequence read coverage files. ### Sickle read trimming We used sickle version 1.33 to trim paired-end reads, with parameters `-q 20 -l 20` to filter reads with quality threshold 20, and requiring reads at least 20 base length. ### Cutadapt adapter trimming We used cutadapt version 1.8.1 with parameter `-m 20` and with standard Illumina Truseq adapter sequence `AGATCGGAAGAGC`. ### Salmon quantitation We used salmon version 0.9.1 for transcript quantitation, using Gencode vM12 comprehensive GTF file, using a index with kmer size 31, and the cutadapt adapter-trimmed reads, with seqBias, VBO, and gcBias options enabled. ### STAR sequence alignment We used STAR version 2.5.1b to align versus Gencode vM12 comprehensive GTF, with the following parameters, based upon ENCODE recommended RNA-seq parameters: ``` --outFilterType Normal --outFilterMultimapNmax 20 --alignSJoverhangMin 8 --alignSJDBoverhangMin 1 --outFilterMismatchNmax 999 --alignIntronMin 20 --alignIntronMax 1000000 --alignMatesGapMax 1000000 --outSAMtype BAM SortedByCoordinate Unsorted --outWigType bedGraph --outWigStrand Stranded --outWigNorm None --outReadsUnmapped Fastx --outMultimapperOrder Random ``` ### featureCounts read counting We used featureCounts version 1.5.1 to produce per-gene counts, using Gencode vM12 transcripts flattened per gene. ## Gene expression analysis summary ### BGA plot, refers to Figure 2 3D between group analysis (BGA) plot of the three biological replicates per cell type per compartment. ```{r bga_plot, dependson="load_gene_data"} bgaExprs <- assays(farrisGeneSE)$counts; bgaExprs[bgaExprs < 7] <- 7; bgaExprsUse <- bgaExprs[rowMins(bgaExprs) < rowMaxs(bgaExprs),,drop=FALSE]; nrow(bgaExprsUse); salmonBga1 <- bga(dataset=bgaExprsUse, classvec=nameVector(colData(farrisGeneSE)$groupName, colnames(farrisGeneSE)), type="coa"); ``` ```{r bga_plotly, dependson=c("bga_plot","color_sub")} salmonBga1ly <- bgaPlotly3d(salmonBga1, axes=c(1,2,3), colorSub=colorSub, useScaledCoords=FALSE, drawVectors="none", drawSampleLabels=FALSE, superGroups=gsub("_.+", "", salmonBga1$fac), ellipseType="none", sceneX=0, sceneY=1, sceneZ=1, verbose=FALSE); salmonBga1ly; ``` ### Comparison with CA1 published data by Nakayama, Ainsley, Cajigas, refers to Figure S2 First we calculate group medians using gene expression data, to determine the set of genes detected above an expression threshold of 128 pseudocounts (log2 = 7). ```{r genes_detected_CA1, fig.height=9, fig.width=6, dependson="load_gene_data"} farrisGeneGroupMedians <- rowGroupMeans(assays(farrisGeneSE)[["counts"]], groups=colData(farrisGeneSE)$groupName, useMedian=TRUE); # Plot histogram of expression in CA1_CB and CA1_DE plotPolygonDensity(farrisGeneGroupMedians[,c("CA1_CB","CA1_DE")], xlim=c(0,20), ylim=c(0,700), ablineV=7); # Detected genes in CA1_CB and CA1_DE CA1detected <- (farrisGeneGroupMedians[,"CA1_CB"] > 7 & farrisGeneGroupMedians[,"CA1_DE"] > 7); CA1detCBandDE <- rownames(farrisGeneGroupMedians)[CA1detected]; length(CA1detCBandDE); ``` Next we load previously gene lists representing detected CA1 dendrite genes from published studies from Nakayama et al, Ainsley et al, and Cajigas et al. These genes are available in the `farrisdata` package, as described above. Produce a 4-way Venn diagram showing the overlapping genes from these studies. ```{r venn_4_way, fig.height=8, fig.width=8, dependson="genes_detected_CA1"} GeneListL <- list( Farris=CA1detCBandDE, Cajigas=CajigasGenes, Ainsley=AinsleyGenes, Nakayama=NakayamaGenes); GeneListIM <- list2im(GeneListL); vps <- limma::vennDiagram(GeneListIM, circle.col=rainbowJam(4)); vps2 <- venn(GeneListL, show.plot=FALSE); ## Retrieve specific overlaps like this: vps2vennL <- attr(vps2, "intersections"); ## vps2vennL[["Farris:Cajigas:Ainsley:Nakayama"]] ``` ### Correlation Centered by Compartment, refers to Figure 3 #### Cell Body Per-sample correlation, centered across all CB samples. First, we use Salmon normalized pseudocounts, restricted to genes where at least one group mean value is above log2(7), which is >= 128 normalized pseudocounts. Then data is centered per **Compartment**, so that CellBody samples are centered by subtracting the mean CellBody expression per gene, and Dendrite samples are centered by subtracting the mean Dendrite expression per gene. This step uses `centerGeneData()` with the argument `centerGroups`. Next we prepare a data.frame with color coding to show the CellType and Compartment values. ```{r sample_corr_cb_prep, dependson=c("load_gene_data","genes_detected_CA1","color_sub")} ## farrisGeneGroupMedians ## Pull out only cell body samples iSamples <- colnames(farrisGeneSE); iSamplesCB <- vigrep("CB", iSamples); iSamplesDE <- vigrep("DE", iSamples); #iSamplesGrp <- colnames(iMatrix7grp); iSamplesGrp <- colnames(farrisGeneGroupMedians); iSamplesGrpCB <- vigrep("CB", iSamplesGrp); iSamplesGrpDE <- vigrep("DE", iSamplesGrp); corrCutoff <- 7; genesAboveCutoff <- (rowMaxs(farrisGeneGroupMedians) >= corrCutoff); genesAboveCutoffCB <- (rowMaxs(farrisGeneGroupMedians[,iSamplesGrpCB]) >= corrCutoff); genesAboveCutoffDE <- (rowMaxs(farrisGeneGroupMedians[,iSamplesGrpDE]) >= corrCutoff); genesAboveCutoffBoth <- (genesAboveCutoffCB & genesAboveCutoffDE); iMatrix7 <- assays(farrisGeneSE[genesAboveCutoff,])[["counts"]]; centerGroups <- gsub("^.+(DE|CB).+$", "\\1", iSamples); iMatrix7ctr <- centerGeneData(iMatrix7, centerGroups=centerGroups); iMatrix7ctrCor <- cor(iMatrix7ctr); ## Generate some color bars to annotate the heatmap colDataColorsRepL <- as.data.frame(colData(farrisGeneSE)[,c("CellType","Compartment")]); colDataColorsRep <- df2colorSub(colDataColorsRepL, colorSub=colorSub); ``` Sample correlation heatmap ```{r sample_corr_cb_CH, fig.height=8, fig.width=10, dependson=c("sample_corr_cb_prep","color_sub")} pheat_colors <- list(Compartment=colorSub[c("CB","DE")], CellType=colorSub[c("CA1","CA2","CA3","DG")]); pheat_breaks <- warpAroundZero(seq(from=-1, to=1, length.out=51), lens=1); cBR <- circlize::colorRamp2(breaks=pheat_breaks, col=getColorRamp("RdBu_r", n=51)); colHA <- HeatmapAnnotation(df=colDataColorsRepL[iSamplesCB,2:1], show_annotation_name=TRUE, border=TRUE, annotation_legend_param=list( title_gp=gpar(fontsize=12, fontface="bold")), col=pheat_colors); rowA <- rowAnnotation(df=colDataColorsRepL[iSamplesCB,2:1], col=pheat_colors, show_annotation_name=TRUE, border=TRUE, annotation_legend_param=list( title_gp=gpar(fontsize=12, fontface="bold")), show_legend=FALSE); corHM <- Heatmap(iMatrix7ctrCor[iSamplesCB,iSamplesCB], name="Correlation", clustering_method_columns="ward.D", clustering_method_rows="ward.D", column_dend_height=unit(20, "mm"), row_dend_width=unit(20, "mm"), top_annotation=colHA, left_annotation=rowA, border=TRUE, heatmap_legend_param=list( grid_width=unit(8, "mm"), border="black", title_gp=gpar(fontsize=12, fontface="bold"), legend_height=unit(30, "mm")), row_dend_side="left", col=cBR); ComplexHeatmap::draw(corHM, merge_legend=TRUE); ``` #### Cell Body and Dendrite Correlation showing CA2_CB correlates highest with CA2_DE, etc. for each CellType. ```{r corr_by_compartment_prep, dependson=c("load_gene_data","genes_detected_CA1","color_sub")} ## farrisGeneGroupMedians[genesAboveCutoff,] iMatrix7grp <- farrisGeneGroupMedians[genesAboveCutoffCB,]; centerGroupsGrp <- gsub("^.*(DE|CB).*$", "\\1", iSamplesGrp); ## 31oct2018 using mean instead of median iMatrix7grpCtr <- centerGeneData(iMatrix7grp, centerGroups=centerGroupsGrp, mean=TRUE); iMatrix7grpCtrCor <- cor(iMatrix7grpCtr); ## Generate some color bars to annotate the heatmap colDataColorsGrpL <- data.frame(rbindList( strsplit(nameVector(iSamplesGrp), "_"))); colnames(colDataColorsGrpL) <- c("CellType","Compartment"); colDataColorsGrp <- df2colorSub(colDataColorsGrpL, colorSub=colorSub); ``` Sample correlation heatmap ```{r corr_by_compartment_CB, fig.height=8, fig.width=10, dependson="corr_by_compartment_prep"} colHA2 <- HeatmapAnnotation(df=colDataColorsGrpL[iSamplesGrp,2:1], show_annotation_name=TRUE, border=TRUE, annotation_legend_param=list( title_gp=gpar(fontsize=12, fontface="bold")), col=pheat_colors); rowA2 <- rowAnnotation(df=colDataColorsGrpL[iSamplesGrp,2:1], col=pheat_colors, show_annotation_name=TRUE, border=TRUE, annotation_legend_param=list( title_gp=gpar(fontsize=12, fontface="bold")), show_legend=FALSE); corHM2 <- Heatmap(iMatrix7grpCtrCor[iSamplesGrp,iSamplesGrp], name="Correlation", clustering_method_columns="ward.D", clustering_method_rows="ward.D", column_dend_height=unit(20, "mm"), row_dend_width=unit(20, "mm"), top_annotation=colHA2, left_annotation=rowA2, border=TRUE, heatmap_legend_param=list( grid_width=unit(8, "mm"), border="black", title_gp=gpar(fontsize=12, fontface="bold"), legend_height=unit(30, "mm")), row_dend_side="left", col=cBR); ComplexHeatmap::draw(corHM2, merge_legend=TRUE); ``` #### Splom plot, refers to Figures 3 and S4 Scatterplot showing the +1,+1 selection of genes, which helps define the gene lists in the next section. ```{r splom_CA2, fig.height=8, fig.width=8, dependson="corr_by_compartment_prep"} corCols <- c("CA2_DE","CA2_CB"); cutCB <- round(digits=3, log2(1.5)); cutDE <- round(digits=3, log2(1.5)); splomL <- lapply(nameVector(c("CA1","CA2","CA3","DG")), function(k) { corCols <- vigrep(k, colnames(iMatrix7grpCtr)); df1 <- as.data.frame(iMatrix7grpCtr[,corCols]); CBhit <- (abs(df1[,1]) > cutCB) * sign(df1[,1]); DEhit <- (abs(df1[,2]) > cutCB) * sign(df1[,2]); CBhitN <- paste0(k, "_", "CBhit"); DEhitN <- paste0(k, "_", "DEhit"); df1[,CBhitN] <- CBhit; df1[,DEhitN] <- DEhit; df1[,"GeneName"] <- rownames(df1); df1; }); #table(splomL[[1]][,3:4]) splomLsub <- lapply(splomL, function(iDF){ subset(iDF, !iDF[,3] %in% 0 | !iDF[,4] %in% 0) }); splomDF2 <- rbindList(lapply(splomLsub, function(iDF){ iDF[,"CellType"] <- gsub("_.+", "", colnames(iDF)[1]); colnames(iDF) <- c("CB","DE","CBhit","DEhit","GeneName","CellType"); iDF; })); #splomDF2[,"row"] <- ifelse(splomDF2$CellType %in% c("CA1","CA2"), "CA1", "CA3"); splomDF2[,"CBDEhit"] <- paste0(splomDF2$CBhit,":",splomDF2$DEhit); ## Color-code the splom points ccColors <- nameVector(rep("grey", 8), unique(splomDF2$CBDEhit)); ccColors[c("1:1","-1:-1")] <- "orangered3"; ccColors[c("-1:1","1:-1")] <- "grey"; ## Gene labels GeneHighlight <- c("Plch2","Rgs14","Necab2","1700024P16Rik"); splomDF2$label <- ifelse(splomDF2$GeneName %in% GeneHighlight, splomDF2$GeneName, ""); splomDF2[,"CBmaxGroupMean"] <- rowMaxs(farrisGeneGroupMedians[splomDF2$GeneName,iSamplesGrpCB]); splomDF2[,"DEmaxGroupMean"] <- rowMaxs(farrisGeneGroupMedians[splomDF2$GeneName,iSamplesGrpDE]); splomDF2[,"CBdet7"] <- (splomDF2[,"CBmaxGroupMean"] > 7)+0; splomDF2[,"DEdet7"] <- (splomDF2[,"DEmaxGroupMean"] > 7)+0; gg2 <- ggplot(splomDF2, aes(x=CB, y=DE, color=CBDEhit, fill=CBDEhit, GeneName=GeneName)) + geom_point(shape=21, size=2) + geom_vline(xintercept=c(-1,1)*0.585, linetype="dashed", color="grey20") + geom_hline(yintercept=c(-1,1)*0.585, linetype="dashed", color="grey20") + facet_wrap(facets=~CellType) + theme_jam() + scale_fill_manual(values=ccColors) + scale_color_manual(values=makeColorDarker(ccColors)) + ggrepel::geom_label_repel(aes(label=label), force=8, fill="white") + theme(legend.position="none"); gg2; ``` ```{r neuronal_gene_lists} ## anyGenes1 is the full set of all genes with abundance > 1 anyGenes1 <- rownames(farrisGeneGroupMedians)[rowMaxs(farrisGeneGroupMedians) > 1]; ## Filtering rules for CB and DE ## corrCutoff is 7 corFilterRuleCB <- (rowMaxs(farrisGeneGroupMedians[,iSamplesGrpCB]) >= corrCutoff); corFilterRuleDE <- (rowMaxs(farrisGeneGroupMedians[,iSamplesGrpDE]) >= corrCutoff); corFilterRule <- (corFilterRuleCB); DEgenes7 <- rownames(farrisGeneGroupMedians)[corFilterRuleDE]; CBgenes7 <- rownames(farrisGeneGroupMedians)[corFilterRuleCB]; DEgenes7nonCB7 <- setdiff(DEgenes7, CBgenes7); # non-pyramidal CBgenes7nonDE7 <- setdiff(CBgenes7, DEgenes7); # "non-DE genes" DEgenes7andCB7 <- intersect(DEgenes7, CBgenes7); # "DE genes" ## Define neuro gene lists geneListsAll <- list(anyGenes1=anyGenes1, DEandCB=DEgenes7andCB7, # detected CBonly=CBgenes7nonDE7, # detected only CB, not detected DE `non-pyramidal`=DEgenes7nonCB7, `CB1andDE1`=unique(subset(splomDF2, CBDEhit %in% "1:1" & CBdet7 %in% c(1) & DEdet7 %in% c(1))$GeneName) ); #lengths(geneListsAll); geneListsDF <- data.frame(GeneList=names(geneListsAll), GeneCount=format(lengths(geneListsAll), big.mark=",", trim=TRUE)); rownames(geneListsDF) <- NULL; colorSubGeneLists <- nameVector(rainbowJam(length(geneListsAll)), names(geneListsAll)); geneListsDF2 <- kable_coloring(geneListsDF, colorSub=colorSubGeneLists, row_color_by="GeneList", returnType="kable") %>% row_spec(0, background="#DDDDDD") geneListsDF2; ``` ### Differential gene expression using limma In preparation. ## Transcript analysis summary ### Definition of detected transcripts We refer to the function `defineDetectedTx()` in the `jampack` R package. The function takes normalized counts, normalized TPM values, sample group information, and several cutoff values: * **cutoffTxPctMax=10** requires a transcript isoform to be at least 10% of the highest isoform present in the same sample. * **cutoffTxExpr=32** requires a transcript isoform to have at least 32 normalized counts. * **cutoffTxTPMExpr=2** requires a transcript isoform to have at least a TPM value of 2. * All of the above criteria must be met for an isoform to be considered "detected." ```{r, detectedTxTPM} #refreshFunctions("farrisSalmonWWS"); detectedTxTPML <- defineDetectedTx( iMatrixTx=assays(farrisTxSE)[["counts"]], iMatrixTxTPM=assays(farrisTxSE)[["tpm"]], groups=colData(farrisTxSE)$groupName, cutoffTxPctMax=10, cutoffTxExpr=32, cutoffTxTPMExpr=2, zeroAsNA=FALSE, tx2geneDF=renameColumn(rowData(farrisTxSE), from=c("geneSymbol","probeID"), to=c("gene_name","transcript_id")), useMedian=FALSE, verbose=FALSE); detectedTx <- detectedTxTPML$detectedTx; numDetectedTx <- length(detectedTx); detectedGenes <- mixedSort(unique(rowData(farrisTxSE[detectedTx,])$geneSymbol)); numDetectedGenes <- length(detectedGenes); ``` The code above defined **`r format(numDetectedTx, big.mark=",")` detected transcripts** by the given criteria, covering **`r format(numDetectedGenes, big.mark=",")` unique gene symbols**. ### Transcript type per gene list e.g. protein_coding, etc. First, as a point of organizing the transcript-gene associations, we read the Gencode GTF file and produce a `data.frame` with the transcript-gene data. Note: This step downloads the Gencode GTF file for version vM12, then extracts data from that file. Once the `data.frame` is extracted, it is stored in a text file for re-use. ```{r, tx2geneDF} vM12gtf <- "ftp://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_mouse/release_M12/gencode.vM12.annotation.gtf.gz"; vM12gtfBase <- basename(vM12gtf); if (!file.exists(vM12gtfBase)) { curl::curl_download(url=vM12gtf, destfile=vM12gtfBase); } tx2geneFile <- file.path(".", "vM12gtf.tx2geneDF.txt"); if (!file.exists(tx2geneFile)) { tx2geneDF <- makeTx2geneFromGtf(GTF=vM12gtfBase, verbose=FALSE); write.table(file=tx2geneFile, x=tx2geneDF, sep="\t", quote=FALSE, na="", col.names=TRUE, row.names=FALSE); } else { tx2geneDF <- read.table(tx2geneFile, sep="\t", check.names=FALSE, as.is=TRUE, fill=TRUE, quote="\"", allowEscapes=FALSE, comment.char="", header=TRUE, stringsAsFactor=FALSE); } ``` Note the logic to define various transcript types is applied below, after the definition of Neuronal Transcript Lists. #### Preparation of 3'UTR data from Gencode We imported Gencode vM12 comprehensive GTF into R using R Bioconductor GenomicFeatures package, with which we derived 3'UTR regions, using `GenomicFeatures::makeTxDbFromGFF()` and `GenomicFeatures::threeUTRsByTranscript()`, respectively. This step re-uses the Gencode GTF file downloaded above, then converts it to a `"Txdb"` object, which is a SQLite relational database format. This database is saved into a file so it can be recalled without re-creating the file again. ```{r, gencode_vM12_gtf} localDb <- file.path(".", "vM12gtf.txdb"); if (!file.exists(localDb)) { vM12txdb <- GenomicFeatures::makeTxDbFromGFF(vM12gtfBase); AnnotationDbi::saveDb(x=vM12txdb, file=localDb); } else { vM12txdb <- AnnotationDbi::loadDb(file=localDb); } ``` ```{r, txdb_refresh, cache=FALSE} ## This snippet simply confirms the vM12txdb object is valid ## otherwise it loads from the source file. This step is ## necessary to resume a previously cached knitr session. if (!DBI::dbIsValid(AnnotationDbi::dbconn(vM12txdb))) { vM12txdb <- AnnotationDbi::loadDb(file=localDb); printDebug("Refreshed vM12txdb."); printDebug('print(find("vM12txdb"))'); print(find("vM12txdb")); } else { printDebug("Did not refresh vM12txdb."); } ``` ```{r, three_utr_length, dependson=c("gencode_vM12_gtf","tx2geneDF")} ## Check for valid vM12txdb since it is not cached properly by knitr if (!DBI::dbIsValid(dbconn(vM12txdb))) { vM12txdb <- AnnotationDbi::loadDb(file=localDb); } ## Grab three prime UTR into a GRangesList gencode3utr <- GenomicFeatures::threeUTRsByTranscript(vM12txdb, use.names=TRUE); values(gencode3utr@unlistData)[,"transcript_id"] <- rep(names(gencode3utr), lengths(gencode3utr)); txMatch <- match(values(gencode3utr@unlistData)[,"transcript_id"], tx2geneDF$transcript_id); values(gencode3utr@unlistData)[,"gene_id"] <- tx2geneDF[txMatch,"gene_id"]; values(gencode3utr@unlistData)[,"gene_name"] <- tx2geneDF[txMatch,"gene_name"]; values(gencode3utr@unlistData)[,"gene_type"] <- tx2geneDF[txMatch,"gene_type"]; values(gencode3utr@unlistData)[,"transcript_type"] <- tx2geneDF[txMatch,"transcript_type"]; names(gencode3utr@unlistData) <- pasteByRow(values(gencode3utr@unlistData)[,c("transcript_id","exon_rank")]); ## Add flags to tx2geneDF, whether a gene or tx has 3'UTR tx2geneDF[,"GeneHas3UTR"] <- tx2geneDF$gene_id %in% values(gencode3utr@unlistData)[,"gene_id"]; tx2geneDF[,"TxHas3UTR"] <- tx2geneDF$transcript_id %in% values(gencode3utr@unlistData)[,"transcript_id"]; ## Summarize widths of three-prime-utr exons by transcript threePrimeTx <- sum(width(gencode3utr)); GencodeVM12mm10threeUtrLength <- sum(width(gencode3utr)); ## non-mito detectedTx detectedTxMito <- subset(tx2geneDF[match(detectedTx, tx2geneDF$transcript_id),], grepl("^mt-", gene_name))$transcript_id; detectedTxNonMito <- setdiff(detectedTx, detectedTxMito); ``` ### Neuronal Transcript Lists The code below creates transcript lists equivalent to the Neuronal Gene Lists above. The main distinction is that the filtering rules are applied to transcript-level data, as opposed to gene-level summary data above. While a gene may be present in one category, perhaps only a subset of its transcript isoforms may be present in that category. Similarly, individual isoforms from the same gene may be present in multiple distinct neuronal categories. The transcript-level subsets are used to produce the 3-prime UTR and CAI plots in subsequent figures. ```{r, neuronal_tx_lists} ## First calculate per-transcript group expression values farrisTxGroupMedians <- rowGroupMeans(assays(farrisTxSE)[["counts"]], groups=colData(farrisTxSE)$groupName, useMedian=TRUE); farrisTxTPMGroupMedians <- rowGroupMeans(assays(farrisTxSE)[["tpm"]], groups=colData(farrisTxSE)$groupName, useMedian=TRUE); ## Define TPM data matrix ## Note: we add +1 to normalized TPM values, which keeps the minimum ## values above zero. Consequently, we reset values which were already ## zero to zero, so they are not adjusted. iMatrixTxTPM <- assays(farrisTxSE)[["tpm"]] + 1; iMatrixTxTPM[assays(farrisTxSE)[["raw_tpm"]] == 0] <- 0 ## Define cutoffs using CB and DE samples ## "Any" uses permissive cutoffs: ## pseudocounts >= 5, TPM >= 1 ## no requirement for isoforms to be a certain percent the max per gene detectedTxTPManyL <- defineDetectedTx( # iMatrixTx=assays(farrisTxSE[,iSamplesCB])[["counts"]], # iMatrixTxTPM=assays(farrisTxSE[,iSamplesCB])[["tpm"]], # groups=colData(farrisTxSE[,iSamplesCB])$groupName, iMatrixTx=assays(farrisTxSE)[["counts"]], iMatrixTxTPM=assays(farrisTxSE)[["tpm"]], groups=colData(farrisTxSE)$groupName, cutoffTxPctMax=0, cutoffTxExpr=5, cutoffTxTPMExpr=1, tx2geneDF=renameColumn(rowData(farrisTxSE), from=c("geneSymbol","probeID"), to=c("gene_name","transcript_id")), useMedian=FALSE, verbose=FALSE); ## Define cutoffs using cell body (CB) samples ## Requires pseudocounts >= 32, TPM >= 2 ## and isoforms must be >= 10 percent the max per gene detectedTxCBTPML <- defineDetectedTx( iMatrixTx=assays(farrisTxSE[,iSamplesCB])[["counts"]], iMatrixTxTPM=assays(farrisTxSE[,iSamplesCB])[["tpm"]], groups=colData(farrisTxSE[,iSamplesCB])$groupName, cutoffTxPctMax=10, cutoffTxExpr=32, cutoffTxTPMExpr=2, tx2geneDF=renameColumn(rowData(farrisTxSE), from=c("geneSymbol","probeID"), to=c("gene_name","transcript_id")), useMedian=FALSE, verbose=FALSE); detectedTxTPMCB <- detectedTxCBTPML$detectedTx; ## Define cutoffs using cell body (DE) samples ## Requires pseudocounts >= 32, TPM >= 2 ## and isoforms must be >= 10 percent the max per gene detectedTxDETPML <- defineDetectedTx( iMatrixTx=assays(farrisTxSE[,iSamplesDE])[["counts"]], iMatrixTxTPM=assays(farrisTxSE[,iSamplesDE])[["tpm"]], groups=colData(farrisTxSE[,iSamplesDE])$groupName, cutoffTxPctMax=10, cutoffTxExpr=32, cutoffTxTPMExpr=2, tx2geneDF=renameColumn(rowData(farrisTxSE), from=c("geneSymbol","probeID"), to=c("gene_name","transcript_id")), useMedian=FALSE, verbose=FALSE); detectedTxTPMDE <- detectedTxDETPML$detectedTx; ## Center data within Compartment farrisTxTPMGroupMediansCtr2 <- centerGeneData(farrisTxTPMGroupMedians, centerGroups=gsub("^.+_", "", colnames(farrisTxTPMGroupMedians)), mean=TRUE, showGroups=FALSE); ## anyTx1 is the full set of all transcripts with abundance > 1 anyTx1 <- rownames(farrisTxGroupMedians)[rowMaxs(farrisTxGroupMedians) > 1]; ## Filtering rules for CB and DE ## corrCutoff is 7 corFilterRuleTxCB <- (rowMaxs(farrisTxGroupMedians[,iSamplesGrpCB]) >= corrCutoff); corFilterRuleTxDE <- (rowMaxs(farrisTxGroupMedians[,iSamplesGrpDE]) >= corrCutoff); corFilterRuleTx <- (corFilterRuleTxCB); ## Subsets of detected transcripts DEtx7 <- rownames(farrisTxGroupMedians)[corFilterRuleTxDE]; CBtx7 <- rownames(farrisTxGroupMedians)[corFilterRuleTxCB]; DEtx7nonCB7 <- setdiff(DEtx7, CBtx7); # non-pyramidal CBtx7nonDE7 <- setdiff(CBtx7, DEtx7); # "non-DE genes" DEtx7andCB7 <- intersect(DEtx7, CBtx7); # "DE genes" ## Plot based upon TPM or counts splomTxL <- lapply(nameVector(c("CA1","CA2","CA3","DG")), function(k) { corCols <- vigrep(k, colnames(farrisTxTPMGroupMediansCtr2)); #df1 <- as.data.frame(farrisTxTPMGroupMediansCtr2[corFilterRuleTx,corCols]); df1 <- as.data.frame(farrisTxTPMGroupMediansCtr2[detectedTxTPMCB,corCols]); CBhit <- (abs(df1[,1]) > cutCB) * sign(df1[,1]); DEhit <- (abs(df1[,2]) > cutCB) * sign(df1[,2]); CBhitN <- paste0(k, "_", "CBhit"); DEhitN <- paste0(k, "_", "DEhit"); df1[,CBhitN] <- CBhit; df1[,DEhitN] <- DEhit; df1[,"GeneName"] <- rownames(df1); df1; }); ## Remove entries with (0,0) no change in any condition splomTxLsub <- lapply(splomTxL, function(iDF){ subset(iDF, !iDF[,3] %in% 0 | !iDF[,4] %in% 0) }); ## Create a data.frame for ggplot splomTxDF2 <- rbindList(lapply(splomTxLsub, function(iDF){ iDF[,"CellType"] <- gsub("_.+", "", colnames(iDF)[1]); colnames(iDF) <- c("CB","DE","CBhit","DEhit","transcript_id","CellType"); iDF; })); splomTxDF2[,"Gene"] <- tx2geneDF[match(splomTxDF2$transcript_id, tx2geneDF$transcript_id),"gene_name"]; splomTxDF2[,"CBDEhit"] <- paste0(splomTxDF2$CBhit, ":", splomTxDF2$DEhit); ## Add rowMaxs for CB and DE to farrisSplomDF2 splomTxDF2[,"CBmaxGroupMedian"] <- rowMaxs(farrisTxGroupMedians[splomTxDF2$transcript_id, iSamplesGrpCB]); splomTxDF2[,"DEmaxGroupMedian"] <- rowMaxs(farrisTxGroupMedians[splomTxDF2$transcript_id, iSamplesGrpDE]); splomTxDF2[,"CBdet7"] <- (splomTxDF2[,"CBmaxGroupMedian"] > corrCutoff)+0; splomTxDF2[,"DEdet7"] <- (splomTxDF2[,"DEmaxGroupMedian"] > corrCutoff)+0; ## Add TPM detected columns splomTxDF2[,"CBmaxGroupMedianTPM"] <- rowMaxs(farrisTxTPMGroupMedians[splomTxDF2$transcript_id, iSamplesGrpCB]); splomTxDF2[,"DEmaxGroupMedianTPM"] <- rowMaxs(farrisTxTPMGroupMedians[splomTxDF2$transcript_id, iSamplesGrpDE]); splomTxDF2[,"CBdetTPM"] <- (splomTxDF2$transcript_id %in% detectedTxTPMCB)+0; splomTxDF2[,"DEdetTPM"] <- (splomTxDF2$transcript_id %in% detectedTxTPMDE)+0; ## Put it all together into tx lists txListsAll <- list(anyGenes1=anyTx1, DEandCB=DEtx7andCB7, # detected CBonly=CBtx7nonDE7, # detected only CB, not detected DE `non-pyramidal`=DEtx7nonCB7, `CB1andDE1`=unique(subset(splomTxDF2, CBDEhit %in% "1:1" & CBdet7 %in% c(1) & #DEdetTPM %in% c(0) & DEdet7 %in% c(1))$transcript_id) ); ######################################################### ## Same logic as above, using TPM instead of pseudocounts detectedTxTPMany <- detectedTxTPManyL$detectedTx; detectedTxTPMCB <- detectedTxCBTPML$detectedTx; detectedTxTPMDE <- detectedTxDETPML$detectedTx; txListsAllTPM <- list(anyGenes1=detectedTxTPMany, DEandCB=intersect(detectedTxTPMCB, detectedTxTPMDE), CBonly=setdiff(detectedTxTPMCB, detectedTxTPMDE), `non-pyramidal`=setdiff(detectedTxTPMDE, detectedTxTPMCB), `CB1andDE1`=unique(subset(splomTxDF2, CBDEhit %in% "1:1" & CBdetTPM %in% c(1) & DEdetTPM %in% c(1))$transcript_id) ); ``` #### Transcript types by transcript list ```{r, tx_by_type} ncTx2geneDFdetTPM <- subset(tx2geneDF, !gene_type %in% c("protein_coding","TEC","Mt_tRNA","Mt_rRNA") & transcript_id %in% detectedTx); ## nc detectedTx has 1548 rows, originally it had 2657 rows ## Convert to list txListsAllTPMncDet <- lapply(txListsAllTPM, function(i){ intersect(i, ncTx2geneDFdetTPM$transcript_id); }); ## lengths(txListsAllTPMncDet) ## 1436, 747, 381, 420, 94 ## % Tx in txListsAll that are protein_coding and have 3'UTRs cdsTx2geneDFdetTPM <- subset(tx2geneDF, gene_type %in% c("protein_coding") & transcript_id %in% detectedTx); txListsAllTPMcdsDet <- lapply(txListsAllTPM, function(i){ iDF <- subset(cdsTx2geneDFdetTPM, transcript_id %in% i); iDF$transcript_id; }); ## lengths(txListsAllTPMcdsDet) ## 24120, 17104, 3424, 5259, 1541 txListsAllTPMcdsDetHas3utr <- lapply(txListsAllTPM, function(i){ iDF <- subset(cdsTx2geneDFdetTPM, transcript_id %in% i); iV <- table(iDF$TxHas3UTR); c(format(iV, scientific=FALSE, big.mark=","), pct=round(1000*nrow(subset(iDF, TxHas3UTR)) / nrow(iDF))/10); }); #txListsAllTPMcdsDetHas3utr; txListsAllTPMcdsDetHas3utrV <- unlist(lapply(txListsAllTPMcdsDetHas3utr, function(i){ as.numeric(gsub(",", "", i[["TRUE"]])); })); ## Subset txListsAllTPM for detected transcripts txListsAllTPMdet <- lapply(txListsAllTPM, function(i){ intersect(i, detectedTx); }); ## Make a small table with txListsAll counts txListsAllDF <- data.frame(geneListsAll=lengths(geneListsAll), txListsAllTPM=lengths(txListsAllTPM), txListsAllTPMdet=lengths(txListsAllTPMdet), non_coding=lengths(txListsAllTPMncDet), pct_non_coding=lengths(txListsAllTPMncDet)/lengths(txListsAllTPMdet), protein_coding=lengths(txListsAllTPMcdsDet), pct_protein_coding=lengths(txListsAllTPMcdsDet)/lengths(txListsAllTPMdet), protein_coding_with3utr=txListsAllTPMcdsDetHas3utrV, pct_protein_coding_with3utr=txListsAllTPMcdsDetHas3utrV/lengths(txListsAllTPMcdsDet), protein_coding_without3utr=lengths(txListsAllTPMcdsDet)-txListsAllTPMcdsDetHas3utrV ); txListsAllDF$otherNc <- txListsAllDF[,"txListsAllTPMdet"] - (txListsAllDF[,"non_coding"] + txListsAllDF[,"protein_coding"]); txListsAllDFuse <- t(as.matrix(txListsAllDF[,c( "protein_coding_with3utr","protein_coding_without3utr", "non_coding","otherNc" )])); ## Scale to 100% txListsAllDFuseScaled <- t(t(txListsAllDFuse) / colSums(txListsAllDFuse))*100; txListsAllDFuseScaled2 <- melt(txListsAllDFuseScaled); txListsAllDFuseScaled2$Var1 <- factor(txListsAllDFuseScaled2$Var1, levels=unique(provigrep(c("other","noncoding|non.coding","without","with","."), as.character(txListsAllDFuseScaled2$Var1)))); txListsAllDFuseScaled2$group <- pasteByRowOrdered(txListsAllDFuseScaled2[,c("Var2","Var1")]); ``` ### Properties of hippocampus dendritic RNAs, refers to S7 #### Heatmap of high-confidence dendritic, cell body retained, and non-pyramidal RNA lists, refers to S7A A heatmap is used to show the pattern of gene expression among the Neuronal Gene Lists defined above. ```{r, heatmap_genelists345_prep} ## Display only genes from the glWhich <- c("CBonly", "non-pyramidal", "CB1andDE1"); gl4 <- unname(unlist(geneListsAll[glWhich])); ## Make a vector of gene list names, named by gene gl4names <- nameVector(rep(names(geneListsAll[glWhich]), lengths(geneListsAll[glWhich])), gl4); iSamplesGrpO <- mixedSort(iSamplesGrp); ## Use all gene group median values iM4 <- farrisGeneGroupMedians[gl4,iSamplesGrpO]; ## Column annotations using each experimental factor (not used) colHA3 <- HeatmapAnnotation(df=colDataColorsGrpL[iSamplesGrpO,2:1], show_annotation_name=TRUE, col=pheat_colors); ## Column annotations using group name and color colHA3grp <- HeatmapAnnotation(Group=rownames(colDataColorsGrpL[iSamplesGrpO,]), annotation_height=unit(7, "mm"), height=unit(7, "mm"), border=TRUE, annotation_legend_param=list( title_gp=gpar(fontsize=12, fontface="bold")), col=list(Group=colorSub[iSamplesGrpO])); ## Row annotations as colors with no labels rowHA3 <- HeatmapAnnotation(which="row", df=data.frame(List=gl4names[rownames(iM4)]), annotation_width=unit(7, "mm"), width=unit(7, "mm"), col=list(List=colorSubGeneLists[names(geneListsAll)[3:5]])); ## Row annotations with block labels rowHA3B <- rowAnnotation( foo=anno_block(gp=gpar(fill=colorSubGeneLists[sort(names(geneListsAll)[3:5])]), labels=sort(names(geneListsAll)[3:5]), labels_gp=gpar(col="white")) ); ## Define color ramp pheat_expr_breaks <- warpAroundZero(seq(from=-2, to=18, length.out=51), baseline=5, lens=4); range(pheat_expr_breaks); cBRexpr <- circlize::colorRamp2(breaks=pheat_expr_breaks, colors=colorRampPalette(tail(warpRamp(getColorRamp("RdBu_r", n=15), lens=0), 12))(51)); ``` ```{r, heatmap_genelists345, fig.height=10, fig.width=8} ## Create the heatmap splitF <- factor(gl4names[rownames(iM4)], levels=(c("CBonly","non-pyramidal","CB1andDE1"))); glHM <- Heatmap(iM4[,iSamplesGrpO], name="Log2 Counts", show_parent_dend_line=FALSE, cluster_columns=FALSE, show_row_names=FALSE, clustering_method_rows="ward.D2", clustering_distance_rows="euclidean", #clustering_method_rows="complete", #clustering_distance_rows="maximum", column_dend_height=unit(20, "mm"), row_dend_width=unit(20, "mm"), top_annotation=colHA3grp, left_annotation=rowHA3B, split=factor(gl4names[rownames(iM4)], levels=names(geneListsAll)[3:5]), gap=unit(3, "mm"), use_raster=TRUE, border=TRUE, heatmap_legend_param=list( grid_width=unit(8, "mm"), border="black", title_gp=gpar(fontsize=12, fontface="bold"), legend_height=unit(30, "mm")), row_dend_side="left", raster_device="CairoPNG", col=cBRexpr); ComplexHeatmap::draw(glHM + rowHA3B, merge_legend=TRUE); ## Print the number of rows and columns in the heatmap ## (broken in non-interactive sessions) if (1 == 2) { legend(x=grconvertX(0.00, "ndc"), y=grconvertY(0.05, "ndc"), legend=paste( dim(iM4[,iSamplesGrpO]), c("rows", "columns")), cex=0.8, xpd=NA, bg="transparent", box.col="transparent"); } ``` #### Percent breakdown of transcript type by transcript list, refers to S7B A stacked bar chart is used to show the relative abundances of transcript types, for each Neuronal Transcript List. ```{r, tx_list_by_type_key, fig.height=6, fig.width=7} ## Define the colors by type nType <- nrow(txListsAllDFuse); nList <- ncol(txListsAllDFuse); txColorSubL <- split(unname( color2gradient(n=nType, reverseGradient=FALSE, gradientWtFactor=1, rainbowJam(nList))), factor(rep(colnames(txListsAllDFuse), each=nType), levels=colnames(txListsAllDFuse))); txColorSubM <- do.call(cbind, txColorSubL); rownames(txColorSubM) <- rownames(txListsAllDFuse); colnames(txColorSubM) <- NULL; par("mar"=c(4,14,4,2)); imageByColors(txColorSubM, main="Transcript Type Color Key"); ``` ```{r, tx_list_by_type_fig, fig.height=12, fig.width=12} txColorSubV <- nameVector(unlist(txColorSubL), txListsAllDFuseScaled2$group); ggTxType <- ggplot(txListsAllDFuseScaled2, aes(x=Var2, fill=group, color=group)) + geom_bar(aes(weight=value)) + scale_fill_manual(values=txColorSubV) + scale_color_manual(values=makeColorDarker(darkFactor=20, txColorSubV)) + theme_jam(base_size=20) + theme( panel.grid.major.x=element_blank(), panel.grid.minor.x=element_blank()) + ylab("Percent") + xlab("Transcript List") + ggtitle("txListsAllTPM by gene_type and 3'UTR") + theme(legend.position="none"); print(ggTxType); ``` #### Heatmap of hypergeometric enrichment P-values of cell type by RNA list, refers to S7C In preparation. #### Violin plot of annotated 3'UTR lengths in each RNA list, refers to S7D We plotted 3'UTR lengths as violin plots using only detected transcripts based upon TPM criteria described previously. ```{r, utr_length, fig.width=12, fig.height=10, dependson=c("three_utr_length")} ## Subset the txListsAll by those with three prime utr data txLists3utr <- lapply(txListsAll, function(i){ intersect(i, names(GencodeVM12mm10threeUtrLength)); }); ## Pull pre-computed data from the farrisdata package data(GencodeVM12mm10cai); data(GencodeVM12mm10cdsLength); data(GencodeVM12mm10caiCtLow); data(GencodeVM12mm10caiCtBad); ## Create data in format for violin plots using ggplot2 violin3utrDF <- data.frame(check.names=FALSE, stringsAsFactors=FALSE, Tx=unlist(txLists3utr), gene_name=tx2geneDF[match(unlist(txLists3utr), tx2geneDF$transcript_id),"gene_name"], Subset=factor(rep(names(txLists3utr), lengths(txLists3utr)), levels=names(txLists3utr)), detectedTx=ifelse(unlist(txLists3utr) %in% detectedTxNonMito, "Detected Tx", "Undetected Tx"), threeUtrLength=GencodeVM12mm10threeUtrLength[unlist(txLists3utr)], txCai=GencodeVM12mm10cai[unlist(txLists3utr)], cdsLength=rmNA(naValue=0, GencodeVM12mm10cdsLength[unlist(txLists3utr)]) ); ## Now make a version that includes Tx counts in the labels violin3utrDF$SubsetCount <- splicejam::factor2label( pasteByRowOrdered(violin3utrDF[,c("Subset","detectedTx")]), itemName="txs"); levels(violin3utrDF$SubsetCount) <- gsub("_([^(]+) Tx (.+) (txs)", " \\2 \\1 \\3", levels(violin3utrDF$SubsetCount)) colorSubGeneLists2 <- unlist(lapply(names(colorSubGeneLists), function(i){ j <- vigrep(i, levels(violin3utrDF$SubsetCount)); nameVector(rep(colorSubGeneLists[i], length(j)), j); })); ## Note some tricky math, we take the median log length, then exponentiate violin3utrDF$SubsetLabel <- splicejam::factor2label( pasteByRowOrdered(violin3utrDF[,c("Subset","detectedTx")]), types="count", aggFun=function(x,...){2^median(log2(x[x>10]),na.rm=TRUE)}, valuesL=violin3utrDF[,"threeUtrLength",drop=FALSE], itemName="txs"); violin3utrDF$log3utr <- log10(1+violin3utrDF$threeUtrLength); ## Subset for specific neuronal groups violin3utrDFsub <- subset(violin3utrDF, Subset %in% names(txListsAll)[c(3,4,5)]); ## 3 prime UTR length violin plots using ggplot2 gg3utr <- ggplot(data=subset(violin3utrDF, detectedTx %in% "Detected Tx"), aes(x=Subset, y=threeUtrLength)) + scale_y_continuous(trans="log10", breaks=c(10,20,50,100,200,300,500,750,1000,1500,2000,5000,10000,20000,50000), limits=10^c(1,4.35)) + #annotation_logticks(sides="lr") + facet_wrap(~detectedTx) + geom_violin(aes(fill=SubsetCount), scale="area", size=1, alpha=0.8, trim=TRUE, draw_quantiles=c(0.5)) + theme_jam(grid.minor.size=0) + scale_color_manual(values=colorSubGeneLists2) + scale_fill_manual(values=colorSubGeneLists2); gg3utr; ``` ##### Statistical tests comparing three prime UTR lengths ```{r, three_utr_stats} ## Wilcoxon t-test comparing the percent CAI threeUtrCBonly <- subset(violin3utrDF, detectedTx %in% "Detected Tx" & Subset %in% "CBonly")$threeUtrLength; threeUtrNonPyr <- subset(violin3utrDF, detectedTx %in% "Detected Tx" & Subset %in% "non-pyramidal")$threeUtrLength; threeUtrCB1andDE1 <- subset(violin3utrDF, detectedTx %in% "Detected Tx" & Subset %in% "CB1andDE1")$threeUtrLength; ## parametric t-tests, using log10-transformed UTR lengths t35utr <- t.test(log10(threeUtrCBonly), log10(threeUtrCB1andDE1)); t34utr <- t.test(log10(threeUtrCBonly), log10(threeUtrNonPyr)); t45utr <- t.test(log10(threeUtrNonPyr), log10(threeUtrCB1andDE1)); ## Wilcoxon non-parametric t-tests wt35utr <- wilcox.test(threeUtrCBonly, threeUtrCB1andDE1); wt34utr <- wilcox.test(threeUtrCBonly, threeUtrNonPyr); wt45utr <- wilcox.test(threeUtrNonPyr, threeUtrCB1andDE1); TtestStatsUtr <- list(t35utr=t35utr, t34utr=t34utr, t45utr=t45utr, wt35utr=wt35utr, wt34utr=wt34utr, wt45utr=wt45utr); TtestStatsUtrDF <- rbindList(lapply(names(TtestStatsUtr), function(t1){ t <- TtestStatsUtr[[t1]]; data.frame(data.name=t$data.name, p.value=t$p.value, method=t$method); })); ## Print stats table colorSubMethod <- group2colors(levels(TtestStatsUtrDF$method)); TtestStatsUtrDF$p.value <- format(TtestStatsUtrDF$p.value, digits=3, trim=TRUE); TtestStatsUtrDF2 <- kable_coloring(TtestStatsUtrDF, colorSub=colorSubMethod, row_color_by="method", returnType="kable") %>% row_spec(0, background="#DDDDDD") TtestStatsUtrDF2; ``` #### Proximal versus distal 3'UTR detected in each RNA list per compartment, refers to S7E ALE elements are defined by unique 5-prime ends of each 3-prime UTR, which allows for overlapping regions downstream, but protects against a prevalent annotation artifact seen with automated gene transcript models. For example, certain genes have a premature alternative STOP codon annotated early in the CDS, causing some isoforms to annotate nearly the entire transcript as 3-prime UTR. We found that taking the 5-prime end of 3-prime UTR regions was able to condense shared 3-prime UTR regions for transcript isoforms per gene, sufficient for this analysis. The resulting data matrix contains rows of ALE elements per gene. The subsequent violin plot compares the ALE2 expression to ALE1 expression, where ALE1 has been subtracted to yield a log2 fold change. ```{r, tx2ale_prep} ## Check for valid vM12txdb since it is not cached properly by knitr if (!DBI::dbIsValid(dbconn(vM12txdb))) { vM12txdb <- AnnotationDbi::loadDb(file=localDb); } ## Define transcripts with ALE regions formatInt <- function(x,...){format(x, scientific=FALSE, trim=TRUE, big.mark=",", ...);} aleL <- tx2ale( threeUtrGRL=gencode3utr, txdb=vM12txdb, detectedTx=detectedTx, iMatrix=assays(farrisTxSE)[["tpm"]], method="flank", verbose=TRUE, tx2geneDF=tx2geneDF); ## use tx2ale to calculate ALE using count data iMatrixAleCt <- log2(1+ shrinkMatrix( 2^(assays(farrisTxSE[names(aleL$tx2ale),])[["counts"]])-1, groupBy=aleL$tx2ale, shrinkFunc=function(x){sum(x, na.rm=TRUE)}, returnClass="matrix")); ``` We also applied logical filtering to the resulting genes, so the Neuronal Gene Lists defined above would be represented by genes in each region according to these rules: * `anyGenes1` is allowed to contain expression data from all genes in the `anyGenes` list from any region. * `DEandCB` only includes genes where the ALE data is meets the thresholds for CB and DE samples. * `CBonly` only includes genes present in CB samples, and excludes DE samples. * `non-pyramidal` only includes genes present in DE samples, and excludes CB samples. * `CB1andDE1` only includes genes where the ALE data is meets the thresholds for CB and DE samples. Data were filtered so that the aggregate ALE TPM value for either ALE1 or ALE2 for each gene was at least 2. ```{r, ale_analysis, dependson="tx2ale_prep", fig.height=10, fig.width=12} groups <- nameVector(colData(farrisTxSE[,iSamples])$groupName, iSamples); facet_groups <- nameVector(gsub("^.+_", "", groups), names(groups)) ## Filter genes for the appropriate geneLists entry subsetCBDE <- function(iDiffAleTall2ctr) { iDFviolinGeneL <- split(iDiffAleTall2ctr$gene_name, pasteByRowOrdered(iDiffAleTall2ctr[,c("geneList","Region")])); DEandCBsubset1 <- intersect(iDFviolinGeneL[["DEandCB_CB"]], iDFviolinGeneL[["DEandCB_DE"]]); CB1andDE1subset1 <- intersect(iDFviolinGeneL[["CB1andDE1_CB"]], iDFviolinGeneL[["CB1andDE1_DE"]]); iDiffAleTall2ctrUse <- subset(iDiffAleTall2ctr, (geneList %in% "DEandCB" & gene_name %in% DEandCBsubset1) | (geneList %in% "CB1andDE1" & gene_name %in% CB1andDE1subset1) | (geneList %in% "CBonly" & Region %in% "CB") | (geneList %in% "non-pyramidal" & Region %in% "DE") | (geneList %in% c("anyGenes1")) ); #print(table(iDiffAleTall2ctrUse[,c("Region","geneList")])); iDiffAleTall2ctrUse; } ## Create violin plot data iMatrixAleGrp <- rowGroupMeans(aleL$iMatrixAle[,iSamples], useMedian=FALSE, groups=groups); aleViolinL <- ale2violin(aleL$iMatrixAle[,iSamples], geneLists=geneListsAll, maxGroupMeanALE=2, groups=groups, lineAlpha=0.1, subsetFunc=subsetCBDE, facet_name="Region", verbose=FALSE, facet_groups=facet_groups); printDebug("Number of genes represented in each panel (TPM calculations):"); table(aleViolinL$iDiffAleTall2ctrFacet[,c("Region","geneList")])/2; print(aleViolinL$ggAle2); ``` #### Violin plot of the average percentage of codons with CAI <= 0.5 per RNA list, refers to S7F Codon CAI values are loaded via the `farrisdata` R package, and can be calculated using methods described in the `jampack` R package suite. ```{r, codon_usage, fig.width=12, fig.height=10, dependson=c("three_utr_length")} ## codon adaptability index (cai) values are pre-computed and ## available from the farrisdata package #data(GencodeVM12mm10cai); ## Subset transcript lists for entries having CAI data, ## which mostly filters for CDS-encoding Txs txListsCai <- lapply(txListsAll, function(i){ intersect(i, names(GencodeVM12mm10cai)) }); txListsCaiTPM <- lapply(txListsAllTPM, function(i){ intersect(i, names(GencodeVM12mm10cai)) }); #txListsCai <- txListsCaiTPM; ## Create data in format for violin plots using ggplot2 violinCaiDF <- data.frame(check.names=FALSE, stringsAsFactors=FALSE, Tx=unlist(txListsCai), gene_name=tx2geneDF[match(unlist(txListsCai), tx2geneDF$transcript_id),"gene_name"], Subset=factor(rep(names(txListsCai), lengths(txListsCai)), levels=names(txListsCai)), detectedTx=ifelse(unlist(txListsCai) %in% detectedTxNonMito, "Detected Tx", "Undetected Tx"), txCai=GencodeVM12mm10cai[unlist(txListsCai)], cdsLength=GencodeVM12mm10cdsLength[unlist(txListsCai)], ctLowCai=GencodeVM12mm10caiCtLow[unlist(txListsCai)], pctLowCai=(GencodeVM12mm10caiCtLow[unlist(txListsCai)] / GencodeVM12mm10cdsLength[unlist(txListsCai)] *100), ctBadCai=GencodeVM12mm10caiCtBad[unlist(txListsCai)], pctBadCai=(GencodeVM12mm10caiCtBad[unlist(txListsCai)] / GencodeVM12mm10cdsLength[unlist(txListsCai)] *100), txCaiQ1=GencodeVM12mm10caiQ1mean[unlist(txListsCai)] ); ## Now make a version that includes Tx counts in the labels violinCaiDF$SubsetCount <- splicejam::factor2label(violinCaiDF$Subset, itemName="txs"); violinCaiDF$SubsetLabel <- splicejam::factor2label(violinCaiDF$Subset, types="count", aggFun=median, valuesL=violinCaiDF[,"pctLowCai",drop=FALSE], itemName="txs"); colorSubGeneLists3 <- unlist(lapply(names(colorSubGeneLists), function(i){ j <- vigrep(i, c( levels(violinCaiDF$SubsetLabel), levels(violinCaiDF$SubsetCount))); nameVector(rep(colorSubGeneLists[i], length(j)), j); })); ## Subset for specific neuronal groups violinCaiDFsub <- subset(violinCaiDF, Subset %in% names(txListsAll)[c(3,4,5)]); ## CAI violin plots using ggplot2 ggPctLowCai <- ggplot(data=subset(violinCaiDFsub, detectedTx %in% "Detected Tx"), aes(x=Subset, y=pctLowCai)) + #aes(x=Subset, y=txCai)) + facet_wrap(~detectedTx) + geom_violin(aes(fill=SubsetLabel), scale="area", #geom_violin(aes(fill=SubsetCount), scale="area", alpha=0.8, trim=TRUE, draw_quantiles=c(0.5)) + ylim(c(0, 20)) + theme_jam(grid.minor.size=0) + scale_color_manual(values=colorSubGeneLists3) + scale_fill_manual(values=colorSubGeneLists3); ggPctLowCai; ``` ##### Statistical tests comparing gene lists ```{r, cai_stats} ## Wilcoxon t-test comparing the percent CAI caiCBonly <- subset(violinCaiDFsub, detectedTx %in% "Detected Tx" & Subset %in% "CBonly")$pctLowCai; caiNonPyr <- subset(violinCaiDFsub, detectedTx %in% "Detected Tx" & Subset %in% "non-pyramidal")$pctLowCai; caiCB1andDE1 <- subset(violinCaiDFsub, detectedTx %in% "Detected Tx" & Subset %in% "CB1andDE1")$pctLowCai; ## parametric t-tests t35cai <- t.test(caiCBonly, caiCB1andDE1); t34cai <- t.test(caiCBonly, caiNonPyr); t45cai <- t.test(caiNonPyr, caiCB1andDE1); ## Wilcoxon non-parametric t-tests wt35cai <- wilcox.test(caiCBonly, caiCB1andDE1); wt34cai <- wilcox.test(caiCBonly, caiNonPyr); wt45cai <- wilcox.test(caiNonPyr, caiCB1andDE1); TtestStatsCai <- list(t35cai=t35cai, t34cai=t34cai, t45cai=t45cai, wt35cai=wt35cai, wt34cai=wt34cai, wt45cai=wt45cai); TtestStatsCaiDF <- rbindList(lapply(names(TtestStatsCai), function(t1){ t <- TtestStatsCai[[t1]]; data.frame(data.name=t$data.name, p.value=t$p.value, method=t$method); })); ## Print stats table colorSubMethod <- group2colors(levels(TtestStatsCaiDF$method)); #TtestStatsCaiDF$p.value <- sfsmisc::pretty10exp(TtestStatsCaiDF$p.value); TtestStatsCaiDF$p.value <- format(TtestStatsCaiDF$p.value, digits=3, trim=TRUE); TtestStatsCaiDF2 <- kable_coloring(TtestStatsCaiDF, colorSub=colorSubMethod, row_color_by="method", returnType="kable") %>% row_spec(0, background="#DDDDDD") TtestStatsCaiDF2; ``` ### Differential transcript isoform analysis limma Differential isoform tests were performed using the R limma package, and the `diffSplice()` function, using normalized TPM abundances. #### Definition of experiment design and contrast matrices The experimental design and contrast matrices are defined based upon the Limma Users Guide, using zero intercept. All pairwise and two-way contrasts are included, provided each contrast changes only one factor at a time. ```{r, design_matrices} ## Design matrix is defined by the sample groups iGroups <- nameVector( colData(farrisTxSE[,iSamples])$groupName, iSamples); ## Re-order factor levels so CA2 is first iGroups <- factor(iGroups, levels=provigrep(c("CA2", "."), levels(iGroups))); iDC <- groups2contrasts(iGroups, returnDesign=TRUE); iDesign <- iDC$iDesign; ## Alternative "manual" method #iDesign <- stats::model.matrix(~0+iGroups); #colnames(iDesign) <- levels(iGroups); #rownames(iDesign) <- names(iGroups); ## Contrasts are CB-DE iContrasts <- iDC$iContrasts; ## Alternative format, useful for custom contrasts #iContrasts <- limma::makeContrasts( # contrasts=c( # "CA2_DE-CA2_CB", # "CA1_DE-CA1_CB"), # levels=iDesign); ``` #### Differential isoform statistical tests Differential isoforms are determined using the `limma::diffSplice()` function, made conveniently available in the `splicejam::runDiffSplice()` function. ```{r, run_diffSplice, dependson=c("design_matrices","detectedTxTPM")} iMatrixTx <- assays(farrisTxSE[,iSamples])[["tpm"]]; iMatrixTx[iMatrixTx == 0] <- NA; diffSpliceL <- runDiffSplice( iMatrixTx=iMatrixTx, detectedTx=detectedTx, tx2geneDF=tx2geneDF, iDesign=iDesign, iContrasts=iContrasts, cutoffFDR=0.05, cutoffFold=1.5, collapseByGene=TRUE, spliceTest="t", verbose=FALSE, useVoom=FALSE); diffSpliceHitsL <- lapply(diffSpliceL$statsDFs, function(iDF){ hitColname <- head(vigrep("^hit ", colnames(iDF)), 1); as.character(subset(iDF, iDF[[hitColname]] != 0)$gene_name); }); ## diffSplice transcript level diffSpliceTxL <- runDiffSplice( iMatrixTx=iMatrixTx, detectedTx=detectedTx, tx2geneDF=tx2geneDF, iDesign=iDesign, iContrasts=iContrasts, cutoffFDR=0.05, cutoffFold=1.5, collapseByGene=FALSE, spliceTest="t", verbose=FALSE, useVoom=FALSE); diffSpliceHitsTxL <- lapply(diffSpliceTxL$statsDFs, function(iDF){ hitColname <- head(vigrep("^hit ", colnames(iDF)), 1); as.character(subset(iDF, iDF[[hitColname]] != 0)$transcript_id); }); #sdim(diffSpliceHitsTxL); ``` ```{r, tx_hit_summary, dependson="run_diffSplice"} ## Subsets of named contrasts DEcon <- unvigrep("_CB", names(diffSpliceHitsTxL)); CBcon <- unvigrep("_DE", names(diffSpliceHitsTxL)); DEconHits <- length(unique(unlist( diffSpliceHitsL[DEcon]))); CBconHits <- length(unique(unlist( diffSpliceHitsL[CBcon]))); DEconHitsTx <- length(unique(unlist( diffSpliceHitsTxL[DEcon]))); CBconHitsTx <- length(unique(unlist( diffSpliceHitsTxL[CBcon]))); CA1CA2CBcon <- vigrep("CA1.*CA2", unvigrep("DE", names(diffSpliceHitsTxL))); CA1CA2DEcon <- vigrep("CA1.*CA2", unvigrep("CB", names(diffSpliceHitsTxL))); CA1CA2CBconHits <- length(diffSpliceHitsL[[CA1CA2CBcon]]); CA1CA2DEconHits <- length(diffSpliceHitsL[[CA1CA2DEcon]]); CA1CA2CBconHitsTx <- length(diffSpliceHitsTxL[[CA1CA2CBcon]]); CA1CA2DEconHitsTx <- length(diffSpliceHitsTxL[[CA1CA2DEcon]]); CA1CA2Twocon <- vigrep("CA1.*CA2_CB", vigrep("DE", names(diffSpliceHitsTxL))); CA1CA2TwoconHitsTx <- length(diffSpliceHitsTxL[[CA1CA2Twocon]]); CA1CA2TwoconHits <- length(diffSpliceHitsL[[CA1CA2Twocon]]); ``` Statistical hits were defined as having a linear fold change >= 1.5, and a Benjamini Hochberg adjusted P-value <= 0.05. * **`r formatInt(nrow(diffSpliceL$statsDFs[[1]]))`** genes with multiple detected isoforms were tested, which include * **`r formatInt(nrow(diffSpliceTxL$statsDFs[[1]]))`** detected isoforms in these multi-isoform genes. * **`r formatInt(length(unique(unlist(diffSpliceHitsTxL))))`** total unique differential transcript isoforms from * **`r formatInt(length(unique(unlist(diffSpliceHitsL))))`** unique genes across all comparisons. Interestingly, only a minor percentage of isoform hits came from cell body comparisons (e.g. CA1 cell body vs CA2 cell body, **`r formatInt(length(diffSpliceHitsTxL[["CA1_CB-CA2_CB"]]))`** differentially spliced transcript isoforms from **`r formatInt(length(diffSpliceHitsL[["CA1_CB-CA2_CB"]]))`** unique genes), suggesting that mature hippocampal neurons overwhelmingly express the same isoforms in similar proportions for co-expressed genes, albeit with nearly **`r formatInt(ceiling(length(diffSpliceHitsL[["CA1_CB-CA2_CB"]]) / 50)*50)`** exceptions. In contrast, the vast majority of differentially expressed isoform hits came from dendrite comparisons (**`r formatInt(DEconHitsTx)`** differentially spliced transcript isoforms from **`r formatInt(DEconHits)`** unique genes). A subset of these hits overlapped with the two-way comparison (e.g. CA2 cell body to dendrite vs CA1 cell body to dendrite, **`r CA1CA2TwoconHitsTx`** differentially spliced transcript isoforms from **`r CA1CA2TwoconHitsTx`** unique genes), suggesting either cell-type and/or compartment specific splicing. #### Export stats results to Excel The stats results will be exported to Excel, using the `openxlsx` R package, storing each stats table as its own worksheet. ```{r, export_stats_xlsx, dependson="run_diffSplice"} GroupColnames <- c("Compartment","Region"); contrastsDF <- iDC$iContrastNames; colnames(contrastsDF)[1:2] <- GroupColnames; contrastsDFsplit <- gsub("^_", "", jamba::pasteByRow( do.call(cbind, lapply(nameVector(GroupColnames), function(i){ ifelse(lengths(strsplit(as.character(contrastsDF[[i]]), ",")) > 1, i, NA) }) ) ) ); contrastsL <- split(as.character(contrastsDF$contrastName), contrastsDFsplit); con2label <- function(i) { gsub("_$", "", cPaste(sep="_", doSort=FALSE, lapply(i, function(j){ sortSamples(preControlTerms=c("CB","DE"), unique(unlist( strsplit(j, "[-_()]+")))); }) )) } ## Loop through each contrast list and save an Excel file for (i in nameVectorN(contrastsL)) { iXlsx <- paste0("DiffIsoforms_by_", i, "_", jamba::getDate(), ".xlsx"); #printDebug(i, "", ": ", iXlsx, # ", ", length(contrastsL[[i]]), " worksheets"); iConLabels <- con2label(contrastsL[[i]]); iMatch <- match(unique(iConLabels), iConLabels); for (iCon in contrastsL[[i]][iMatch]) { iLab <- con2label(iCon); #printDebug(" ", iLab); #iDFtx <- diffSpliceTxL$statsDFs[[iCon]]; ## Gene-centric stats table iDFgene <- diffSpliceL$statsDFs[[iCon]]; ## Remove t-statistic column, reorder hit column iDFgene <- dplyr::select(iDFgene, -tidyselect::matches("^t ")) %>% dplyr::select(tidyselect::matches("gene|transcript"), tidyselect::matches("^hit "), tidyselect::everything()); jamba::writeOpenxlsx(file=iXlsx, x=iDFgene, sheetName=iLab, append=(!iCon == contrastsL[[i]][[1]]), highlightColumns=igrep("gene|transcript", colnames(iDFgene)), pvalueColumns=igrep("p.val|fdr", colnames(iDFgene)), lfcColumns=igrep("logfc", colnames(iDFgene)), numColumns=igrep("groupmean", colnames(iDFgene)), intColumns=igrep("^numTx", colnames(iDFgene)), intRule=c(0,7,15), hitColumns=igrep("^hit ", colnames(iDFgene)), freezePaneColumn=max(igrep("gene|transcript", colnames(iDFgene))), freezePaneRow=2 ); } } ``` ### Splicing, Sashimi plots Two genes with interesting alternative transcript isoforms are **Ntrk2** and **Gria1**. Sashimi plots are shown below, which represent the RNA-seq sequence read coverage across the exons, and the splice junction reads that span from exon to exon. The plots also include the transcript-exon model, and the flattened gene-exon model. First, some Sashimi plot data is prepared: * Exons per transcript, and CDS exons per transcript, which are derived from the Gencode GTF loaded previously. * `filesDF` which is a `data.frame` describing each RNA-seq file used to load sequence read coverage, and splice junction reads. * `sample_id` which is a subset of sample_id values described in `filesDF`. ```{r, sashimi_prep} ## Check for valid TxDb if (!DBI::dbIsValid(AnnotationDbi::dbconn(vM12txdb))) { vM12txdb <- AnnotationDbi::loadDb(file=localDb); } ## Obtain the exons per transcript exonsByTx <- exonsBy(vM12txdb, by="tx", use.names=TRUE); ## Obtain CDS exons per transcript, used to show UTR as non-CDS cdsByTx <- cdsBy(vM12txdb, by="tx", use.names=TRUE); ## flatten exons by gene for these two genes flatExonsByGene2 <- flattenExonsBy(exonsByTx=exonsByTx, tx2geneDF=tx2geneDF, detectedTx=detectedTx, cdsByTx=cdsByTx, by="gene", genes=c("Ntrk2", "Gria1")); ## flatten CDS exons by gene for these two genes flatExonsByTx2 <- flattenExonsBy(exonsByTx=exonsByTx, tx2geneDF=tx2geneDF, detectedTx=detectedTx, cdsByTx=cdsByTx, by="tx", genes=c("Ntrk2", "Gria1")); ## Define the files used for the sashimi plots filesDF <- farrisdata::farris_sashimi_files_df; ## Define a subset of sample_id to display sample_id <- c("CA1_CB", "CA1_DE", "CA2_CB", "CA2_DE"); ``` #### Sashimi plot for Ntrk2 ```{r, sashimi_ntrk2_prep, fig.height=14, fig.width=16} ## prepare sashimi plot data for Ntrk2 sashimi_ntrk2 <- prepareSashimi(gene="Ntrk2", tx2geneDF=tx2geneDF, flatExonsByTx=flatExonsByTx2, flatExonsByGene=flatExonsByGene2, colorSub=farrisdata::colorSub, filesDF=filesDF, sample_id=sample_id); ``` ```{r, sashimi_ntrk2, fig.height=14, fig.width=16} ## prepare gene-exon model for Ntrk2 gg_ntrk2 <- gene2gg(gene="Ntrk2", gene_order="last", labelExons=FALSE, exonLabelSize=6, direction="y", flatExonsByGene=flatExonsByGene2, flatExonsByTx=flatExonsByTx2) ## use cowplot to assemble sashimi plot and gene-exon model cp_ntrk2 <- cowplot::plot_grid( gg_ntrk2 + theme(axis.text.x=element_blank()) + #ggtitle(NULL) + xlab(NULL), plotSashimi(sashimi_ntrk2, color_sub=farrisdata::colorSub), ncol=1, align="v", axis="lr", rel_heights=c(1,2)); cp_ntrk2; ``` #### Sashimi plot for Gria1 ```{r, sashimi_gria1_prep, fig.height=14, fig.width=16} ## prepare sashimi plot data for Ntrk2 sashimi_gria1 <- prepareSashimi(gene="Gria1", tx2geneDF=tx2geneDF, flatExonsByTx=flatExonsByTx2, flatExonsByGene=flatExonsByGene2, filesDF=filesDF, minJunctionScore=200, sample_id=sample_id); ``` ```{r, sashimi_gria1, fig.height=14, fig.width=16} ## prepare gene-exon model for Ntrk2 gg_gria1 <- gene2gg(gene="Gria1", gene_order="last", labelExons=FALSE, exonLabelSize=6, direction="y", flatExonsByGene=flatExonsByGene2, flatExonsByTx=flatExonsByTx2) ## use cowplot to assemble sashimi plot and gene-exon model cp_gria1 <- cowplot::plot_grid( gg_gria1 + theme(axis.text.x=element_blank()) + #ggtitle(NULL) + xlab(NULL), plotSashimi(sashimi_gria1, color_sub=farrisdata::colorSub), ncol=1, align="v", axis="lr", rel_heights=c(1,2)); cp_gria1; ``` Zoom into the differential slice event, that switched from the "flop" to the "flip" form of Gria1: ```{r, sashimi_gria1_zoom, fig.height=10, fig.width=10} ## use cowplot to assemble sashimi plot and gene-exon model gria1_r <- ranges(range(subset(flatExonsByGene2[["Gria1"]], gene_nameExon %in% c("Gria1_exon13", "Gria1_exon17a")))) gria1_zoom <- c(start(gria1_r), end(gria1_r)); cp_gria1_zoom <- cowplot::plot_grid( gg_gria1 + theme(axis.text.x=element_blank()) + coord_cartesian(xlim=gria1_zoom, expand=FALSE) + xlab(NULL), plotSashimi(sashimi_gria1, label_coords=gria1_zoom, color_sub=farrisdata::colorSub) + coord_cartesian(xlim=gria1_zoom, expand=FALSE), ncol=1, align="v", axis="lr", rel_heights=c(1,3)); cp_gria1_zoom; ``` ## Gene Set Enrichment ### Selection of gene hits Used DESeq-normalized expression values, tested versus MSigDB v6.0 using hypergeometric enrichment via the R function `stats::phyper()`. A `data.frame` summary of statistical results was prepared, which included: * pathway name * pathway "Source" and "Category" as defined by MSigDB * number of genes per pathway * number of gene hits found per pathway * hypergeometric enrichment P-value * FDR-adjusted P-value * comma-delimited gene hits per pathway The gene symbols obtained from MSigDB were converted to most recent mouse Entrez gene symbols using the Bioconductor package `"org.Mm.eg.db"`. This same process was used with the gene symbols from the RNA-seq analysis steps above, in order to ensure the gene symbols in MSigDB and RNA-seq analysis were using the same comparable version of Entrez gene symbols. One `data.frame` result was prepared for each gene hit list, and was further analyzed using methods in the `"multienrichjam"` package. ### MultiEnrichMap MultiEnrichMap is an analysis workflow that extends previous innovative analysis workflows: * "[EnrichmentMap](https://www.baderlab.org/Software/EnrichmentMap)" by the Bader lab, ref: [Merico,et al, PLoS One, 2010](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0013984)) * "[clusterProfiler](https://bioconductor.org/packages/release/bioc/html/clusterProfiler.html)" and "[enrichplot](https://bioconductor.org/packages/release/bioc/html/enrichplot.html)" by Dr. Guangchuang Yu. The previous EnrichmentMap and EnrichMap workflows are intended to help understand and visualize gene set enrichment results in the context of one gene hit list. MultiEnrichMap is an extension of those workflow that compares multiple enrichment results. This workflow is implemented in an R package "multienrichjam", which can be installed with > `devtools::install_github("jmw86069/multienrichjam")` Subset pathways for top 15 per source-category Heatmap of enrichment P-values. Cnet plot