3_Enrichment_MutantVsWT.Rmd

---
title: "Enrichment Analysis: Mutant Vs Wild-Type"
author: "Steve Pederson"
date: "`r format(Sys.time(), '%d %B, %Y')`"
output: workflowr::wflow_html
editor_options:
  chunk_output_type: console
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(
    autodep = TRUE,
    echo = TRUE,
    warning = FALSE,
    message = FALSE,
    fig.align = "center"
)
```


# Setup

```{r loadPackages}
library(tidyverse)
library(magrittr)
library(edgeR)
library(scales)
library(pander)
library(goseq)
library(msigdbr)
library(AnnotationDbi)
library(RColorBrewer)
library(ngsReports)
library(UpSetR)
library(pheatmap)
```

```{r setOpts}
theme_set(theme_bw())
panderOptions("table.split.table", Inf)
panderOptions("table.style", "rmarkdown")
panderOptions("big.mark", ",")
```



```{r samplesAndLabels}
samples <- read_csv("data/samples.csv") %>%
  distinct(sampleName, .keep_all = TRUE) %>%
  dplyr::select(sample = sampleName, sampleID, genotype) %>%
  mutate(
    genotype = factor(genotype, levels = c("WT", "Het", "Hom")),
    mutant = genotype %in% c("Het", "Hom"),
    homozygous = genotype == "Hom"
  )
genoCols <- samples$genotype %>%
  levels() %>%
  length() %>%
  brewer.pal("Set1") %>%
  setNames(levels(samples$genotype))
```


```{r loadFits}
dgeList <- here::here("data/dgeList.rds") %>% read_rds()
cpmPostNorm <- here::here("data/cpmPostNorm.rds") %>% read_rds()
entrezGenes <- dgeList$genes %>%
  dplyr::filter(!is.na(entrezid)) %>%
  unnest(entrezid) %>%
  dplyr::rename(entrez_gene = entrezid)
```

```{r topTables}
deTable <- here::here("output", "psen2VsWT.csv") %>% 
  read_csv() %>%
  mutate(
    entrezid = dgeList$genes$entrezid[gene_id]
  )
```

```{r formatP}
formatP <- function(p, m = 0.0001){
  out <- rep("", length(p))
  out[p < m] <- sprintf("%.2e", p[p<m])
  out[p >= m] <- sprintf("%.4f", p[p>=m])
  out
}
```


# Introduction

Enrichment analysis for this dataset present some challenges.
Despite normalisation to account for gene length and GC bias, some appeared to still be present in the final results.
In addition, the confounding of incomplete rRNA removal with genotype may lead to other distortions in both DE genes and ranking statistics.

Two steps for enrichment analysis will be undertaken.

1. Testing for enrichment within *discrete sets of DE genes* as defined in the previous steps
2. Testing for enrichment using the complete set of results using `fry`.

Testing for enrichment *within discrete gene sets* will be performed using `goseq` as this allows for the incorporation of a single covariate as a predictor of differential expression.
GC content, gene length and correlation with rRNA removal can all be supplied as separate covariates.

For enrichment within larger gene lists,  `fry` can take into account inter-gene correlations. 
Values supplied will be logCPM for each gene/sample after being adjusted for GC and length biases.


# Databases used for testing

Data was sourced using the `msigdbr` package.
The initial database used for testing was the Hallmark Gene Sets, with mappings from gene-set to EntrezGene IDs performed by the package authors. 

## Hallmark Gene Sets

```{r hm}
hm <- msigdbr("Danio rerio", category = "H")  %>% 
  left_join(entrezGenes) %>%
  dplyr::filter(!is.na(gene_id)) %>%
  distinct(gs_name, gene_id, .keep_all = TRUE)
hmByGene <- hm %>%
  split(f = .$gene_id) %>%
  lapply(extract2, "gs_name")
hmByID <- hm %>%
  split(f = .$gs_name) %>%
  lapply(extract2, "gene_id")
```

Mappings are required from gene to pathway, and Ensembl identifiers were used to map from gene to pathway, based on the mappings in the previously used annotations (Ensembl Release 98).
A total of `r comma(length(hmByGene))` Ensembl IDs were mapped to pathways from the hallmark gene sets.

## KEGG Gene Sets


```{r kg}
kg <- msigdbr("Danio rerio", category = "C2", subcategory = "CP:KEGG")  %>% 
  left_join(entrezGenes) %>%
  dplyr::filter(!is.na(gene_id)) %>%
  distinct(gs_name, gene_id, .keep_all = TRUE)
kgByGene <- kg  %>%
  split(f = .$gene_id) %>%
  lapply(extract2, "gs_name")
kgByID <- kg  %>%
  split(f = .$gs_name) %>%
  lapply(extract2, "gene_id")
```

The same mapping process was applied to KEGG gene sets.
A total of `r comma(length(kgByGene))` Ensembl IDs were mapped to pathways from the KEGG gene sets.

## Gene Ontology Gene Sets

```{r goSummaries}
goSummaries <- url("https://uofabioinformaticshub.github.io/summaries2GO/data/goSummaries.RDS") %>%
  readRDS() %>%
  mutate(
    Term = Term(id),
    gs_name = Term %>% str_to_upper() %>% str_replace_all("[ -]", "_"),
    gs_name = paste0("GO_", gs_name)
    )
minPath <- 3
```


```{r go}
go <- msigdbr("Danio rerio", category = "C5") %>% 
  left_join(entrezGenes) %>%
  dplyr::filter(!is.na(gene_id)) %>%
  left_join(goSummaries) %>% 
  dplyr::filter(shortest_path >= minPath) %>%
  distinct(gs_name, gene_id, .keep_all = TRUE)
goByGene <- go %>%
  split(f = .$gene_id) %>%
  lapply(extract2, "gs_name")
goByID <- go %>%
  split(f = .$gs_name) %>%
  lapply(extract2, "gene_id")
```


For analysis of gene-sets from the GO database, gene-sets were restricted to those with `r minPath` or more steps back to the ontology root terms.
A total of `r comma(length(goByGene))` Ensembl IDs were mapped to pathways from restricted database of `r comma(length(goByID))` GO gene sets.

```{r gsSizes}
gsSizes <- bind_rows(hm, kg, go) %>% 
  dplyr::select(gs_name, gene_symbol, gene_id) %>% 
  chop(c(gene_symbol, gene_id)) %>%
  mutate(
    gs_size = vapply(gene_symbol, length, integer(1)),
    de_id = lapply(
      X = gene_id, 
      FUN = intersect, 
      y = dplyr::filter(deTable, DE)$gene_id
      ),
    de_size = vapply(de_id, length, integer(1))
  )
```


# Enrichment in the DE Gene Set


The first step of analysis using `goseq`, regardless of the gene-set, is estimation of the probability weight function (PWF) which quantifies the probability of a gene being considered as DE based on a single covariate.
As GC content and length should have been accounted for during conditional-quantile normalisation, these were not required for any bias.
However, the *gene-level correlations with rRNA contamination* were instead used a predictor of bias in selection of a gene as being DE.


```{r rawFqc}
rawFqc <- list.files(
  path = here::here("data/0_rawData/FastQC/"),
  pattern = "zip",
  full.names = TRUE
) %>%
  FastqcDataList()
gc <- getModule(rawFqc, "Per_sequence_GC") %>%
  group_by(Filename) %>% 
  mutate(Freq = Count / sum(Count)) %>%
  ungroup()
gcDev <- gc %>%
  left_join(getGC(gcTheoretical, "Drerio", "Trans")) %>%
  mutate(
    sample = str_remove(Filename, "_R[12].fastq.gz"),
    resid = Freq - Drerio
  ) %>% 
  left_join(samples) %>%
  group_by(sample) %>%
  summarise(
    ss = sum(resid^2),
    n = n(),
    sd = sqrt(ss / (n - 1))
  )
```

```{r riboCors}
riboVec <- structure(gcDev$sd, names = gcDev$sample)
riboCors <- cpm(dgeList, log = TRUE) %>%
  apply(1, function(x){
    cor(x, riboVec[names(x)])
  })
```

Values were calculated as per the previous steps, using the logCPM values for each gene, with the sample-level standard deviations from the theoretical GC distribution being used as a measure of rRNA contamination.
For estimation of the probability weight function, squared correlations were used to place negative and positive correlations on the same scale.
This accounts for genes which are both negatively & positively biased by the presence of excessive rRNA.
Clearly, the confounding of genotype with rRNA means that some genes driven by the genuine biology may be down-weighted under this approach.


```{r riboPwf, fig.height=4, fig.width=4, fig.cap="*Using this approach, it was clear that correlation with rRNA proportions significantly biased the probability of a gene being considered as DE.*"}
riboPwf <- deTable %>%
  mutate(riboCors = riboCors[gene_id]^2) %>%
  dplyr::select(gene_id, DE, riboCors) %>%
  distinct(gene_id, .keep_all = TRUE) %>% 
  with(
    nullp(
      DEgenes = structure(
        as.integer(DE), names = gene_id
      ), 
      genome = "danRer10", 
      id = "ensGene", 
      bias.data = riboCors,
      plot.fit = FALSE
    )
  )
plotPWF(riboPwf, main = "Bias from rRNA proportions")
```

All gene-sets were then tested using this PWF.

## Hallmark Gene Sets

```{r hmRiboGoSeq}
hmRiboGoseq <- goseq(riboPwf, gene2cat = hmByGene) %>%
  as_tibble %>%
  dplyr::filter(numDEInCat > 0) %>%
  mutate( 
    adjP = p.adjust(over_represented_pvalue, method = "bonf"),
    FDR = p.adjust(over_represented_pvalue, method = "fdr")
  ) %>%
  dplyr::select(-contains("under")) %>%
  dplyr::rename(
    gs_name = category,
    PValue = over_represented_pvalue
  )
```

No gene-sets achieved significance in the DE genes with the lowest FDR being `r percent(min(hmRiboGoseq$FDR))`

## KEGG Gene Sets


```{r kgRiboGoseq}
kgRiboGoseq <- goseq(riboPwf, gene2cat = kgByGene) %>%
  as_tibble %>%
  dplyr::filter(numDEInCat > 0) %>%
  mutate( 
    adjP = p.adjust(over_represented_pvalue, method = "bonf"),
    FDR = p.adjust(over_represented_pvalue, method = "fdr")
  ) %>%
  dplyr::select(-contains("under"))  %>%
  dplyr::rename(
    gs_name = category,
    PValue = over_represented_pvalue
  )
```

```{r kgRiboGoseqTable}
kgRiboGoseq %>%
  dplyr::slice(1:5) %>%
  mutate(
    p = formatP(PValue),
    adjP = formatP(adjP),
    FDR = formatP(FDR)
  ) %>%
  dplyr::select(
    `Gene Set` = gs_name,
    DE = numDEInCat,
    `Set Size` = numInCat,
    PValue,
    `p~bonf~` = adjP,
    `p~FDR~` = FDR
  ) %>%
  pander(
    justify = "lrrrrr",
    caption = paste(
      "The", nrow(.), "most highly-ranked KEGG pathways.",
      "Bonferroni-adjusted p-values are the most stringent and give high",
      "confidence when below 0.05."
    )
  )
```

Notably, the KEGG gene-set for Ribosomal genes was detected as enriched in the set of DE genes, with no other KEGG gene-sets being considered significant.

## GO Gene Sets

```{r goRiboGoseq}
goRiboGoseq <- goseq(riboPwf, gene2cat = goByGene) %>%
  as_tibble %>%
  dplyr::filter(numDEInCat > 0) %>%
  mutate( 
    adjP = p.adjust(over_represented_pvalue, method = "bonf"),
    FDR = p.adjust(over_represented_pvalue, method = "fdr")
  ) %>%
  dplyr::select(-contains("under")) %>%
  dplyr::rename(
    gs_name = category,
    PValue = over_represented_pvalue
  )
```

```{r goRiboGoseqTable}
goRiboGoseq %>%
  dplyr::filter(adjP < 0.05) %>%
  mutate(
    p = formatP(PValue),
    adjP = formatP(adjP),
    FDR = formatP(FDR)
  ) %>%
  dplyr::select(
    `Gene Set` = gs_name,
    DE = numDEInCat,
    `Set Size` = numInCat,
    PValue,
    `p~bonf~` = adjP,
    `p~FDR~` = FDR
  ) %>%
  pander(
    justify = "lrrrrr",
    caption = paste(
      "*The", nrow(.), "most highly-ranked GO terms.",
      "Bonferroni-adjusted p-values are the most stringent and give high",
      "confidence when below 0.05, with all terms here reaching this threshold.",
      "However, most terms once again indicate the presence of rRNA.*"
    )
  )
```

# Enrichment Testing on the Complete Set of Genes

## Hallmark Gene Sets

```{r hmFry}
hmFry <- cpmPostNorm %>%
  fry(
    index = hmByID,
    design = dgeList$design,
    contrast = "mutant",
    sort = "directional"
    ) %>%
  rownames_to_column("gs_name") %>%
  as_tibble()
```

```{r}
hmFry %>%
  dplyr::filter(
    PValue < 0.01
  ) %>%
  left_join(gsSizes) %>%
  mutate(gs_name = str_remove(gs_name, "HALLMARK_")) %>%
  dplyr::select(
    `Hallmark Gene Set` = gs_name,
    `Expressed Genes` = NGenes,
    `DE Genes` = de_size,
    Direction,
    PValue, FDR
    ) %>%
  pander(
    justify = "lrrlrr",
    caption = "Enriched Hallmark Gene Sets, using a threshold of p < 0.01, which corresponds to an FDR of 0.02"
  )
```


### Exploration of significant Hallmark Gene Sets

```{r hmFryUpset, fig.width=10, fig.cap = "*UpSet plot indicating distribution of DE genes within all significant HALLMARK gene sets. Gene sets were restricted to those with an FDR < 0.05 and at least 5 DE genes. The plot is truncated at the right for simplicity. Most gene sets seem relatively independent of each other with regard to DE genes.*"}
hmFry %>% 
  left_join(gsSizes) %>%
  dplyr::filter(de_size >= 5, FDR < 0.05) %>%
  dplyr::select(gs_name, de_id) %>%
  unnest(de_id) %>% 
  mutate(
    gs_name = str_remove(gs_name, "HALLMARK_"),
    gs_name = fct_lump(gs_name, n = 14)
    ) %>% 
  split(f = .$gs_name) %>% 
  lapply(magrittr::extract2, "de_id") %>% 
  fromList() %>% 
  upset(
    nsets = length(.), 
    nintersects = 20,
    order.by = "freq", 
    mb.ratio = c(0.6, 0.4),
    sets.x.label = "Number Of DE Genes"
    )
```

```{r hmHeat, fig.height=9, fig.width=10, fig.cap = "*Gene expression patterns for all DE genes in HALLMARK gene sets, containing more than 12 DE genes.*"}
hmHeat <- hmFry %>% 
  left_join(gsSizes) %>%
  dplyr::filter(de_size >= 12, FDR < 0.05) %>%
  dplyr::select(gs_name, de_id) %>%
  unnest(de_id) %>%
  left_join(dgeList$genes, by = c("de_id" = "gene_id")) %>%
  dplyr::select(gs_name, de_id, gene_name) %>%
  mutate(belongs = TRUE) %>%
  pivot_wider(
    id_cols = c(de_id, gene_name),
    names_from = gs_name,
    values_from = belongs,
    values_fill = list(belongs = FALSE)
  ) %>%
  left_join(
    cpmPostNorm %>%
      as.data.frame() %>%
      rownames_to_column("de_id")
  ) 
hmHeat %>%
  dplyr::select(gene_name, starts_with("Ps")) %>%
  as.data.frame() %>%
  column_to_rownames("gene_name") %>%
  as.matrix() %>%
  pheatmap(
    color = viridis_pal(option = "magma")(100),
    legend_breaks = c(seq(-2, 8, by = 2), max(.)),
    legend_labels = c(seq(-2, 8, by = 2), "logCPM\n"),
    labels_col = hmHeat %>%
      dplyr::select(starts_with("Ps")) %>% 
      colnames() %>% 
      enframe(name = NULL) %>% 
      dplyr::rename(sample = value) %>%
      left_join(dgeList$samples) %>%
      dplyr::select(sample, sampleID) %>%
      with(
        structure(sampleID, names = sample)
      ),
    cutree_rows = 6,
    cutree_cols = 2,
    annotation_names_row = FALSE,
    annotation_row = hmHeat %>%
      dplyr::select(gene_name, starts_with("HALLMARK_")) %>%
      mutate_at(vars(starts_with("HALL")), as.character) %>%
      as.data.frame() %>%
      column_to_rownames("gene_name") %>%
      set_colnames(
        str_remove(colnames(.), "HALLMARK_")
      )
  )
```


## KEGG Gene Sets

```{r kgFry}
kgFry <- cpmPostNorm %>%
  fry(
    index = kgByID,
    design = dgeList$design,
    contrast = "mutant",
    sort = "directional"
    ) %>%
  rownames_to_column("gs_name") %>%
  as_tibble()
```



```{r}
kgFry %>%
  dplyr::filter(
    PValue < 0.005
  ) %>%
  left_join(gsSizes) %>%
  mutate(gs_name = str_remove(gs_name, "HALLMARK_")) %>%
  dplyr::select(
    `KEGG Gene Set` = gs_name,
    `Expressed Genes` = NGenes,
    `DE Genes` = de_size,
    Direction,
    PValue, FDR
    ) %>%
  pander(
    justify = "lrrlrr",
    caption = "Enriched KEGG Gene Sets, using a threshold p < 0.005, which corresponds to an FDR of 0.02"
  )
```

### Exploration of Significant KEGG Gene Sets

```{r kgFryUpset, fig.width=8, fig.cap = "*UpSet plot indicating distribution of DE genes within all significant terms from the KEGG gene sets. Gene sets were restricted to those with an FDR < 0.05 and four or more DE genes. There is considerable overlap between DE genes in Oxidative Phosphorylation and both Parkinson's and Huntington's Disease, indicating that these terms essentially capture the same expression signal within this dataset.*"}
kgFry %>% 
  left_join(gsSizes) %>%
  dplyr::filter(de_size >= 4, FDR < 0.05) %>%
  dplyr::select(gs_name, de_id) %>%
  unnest(de_id) %>% 
  mutate(
    gs_name = str_remove(gs_name, "KEGG_"),
    gs_name = fct_lump(gs_name, n = 14)
    ) %>% 
  split(f = .$gs_name) %>% 
  lapply(magrittr::extract2, "de_id") %>% 
  fromList() %>% 
  upset(
    nsets = length(.), 
    nintersects = 20,
    order.by = "freq", 
    mb.ratio = c(0.6, 0.4),
    sets.x.label = "Number Of DE Genes"
    )
```


```{r riboOxHeatmap, fig.width=8, fig.height=8, fig.cap = "*Given that the largest gene sets within the KEGG results were Oxidative Phosphorylation and Ribosomal gene sets, all DE genes associated with these KEGG terms are displayed. All genes appear to show increased expression in mutant samples.*"}
riboOx <- kg %>%
  dplyr::filter(
    grepl("OXIDATIVE_PHOS", gs_name) | grepl("RIBOSOME", gs_name),
    gene_id %in% dplyr::filter(deTable, DE)$gene_id
    ) %>%
  mutate(gs_name = str_remove(gs_name, "(KEGG|HALLMARK)_")) %>%
  distinct(gs_name, gene_id) %>%
  left_join(
    cpmPostNorm %>% 
      as.data.frame() %>% 
      rownames_to_column("gene_id") %>%
      as_tibble()
    ) %>%
  pivot_longer(cols = starts_with("Ps"), names_to = "sample", values_to = "CPM") %>%
  left_join(dgeList$samples) %>% 
  left_join(dgeList$genes) %>%
  dplyr::select(gs_name, gene_name, sampleID, genotype, CPM) %>%
  pivot_wider(
    id_cols = c(gs_name, gene_name),
    values_from = CPM,
    names_from = sampleID
  ) 
riboOx %>% 
  dplyr::select(-gs_name) %>%
  as.data.frame() %>%
  column_to_rownames("gene_name") %>%
  as.matrix() %>%
  pheatmap(
    annotation_row = data.frame(
      GeneSet = riboOx$gs_name %>% str_replace("OXI.+", "OXPHOS"), 
      row.names = riboOx$gene_name
      ),
    color = viridis_pal(option = "magma")(100),
    legend_breaks = c(seq(-2, 8, by = 2), max(.)),
    legend_labels = c(seq(-2, 8, by = 2), "logCPM\n"),
    annotation_names_row = FALSE,
    cutree_rows = 3,
    cutree_cols = 2
  )
```


## GO Gene Sets

```{r goFry}
goFry <- cpmPostNorm %>%
  fry(
    index = goByID,
    design = dgeList$design,
    contrast = "mutant",
    sort = "directional"
    ) %>%
  rownames_to_column("gs_name") %>%
  as_tibble()
```

```{r}
goFry %>%
  left_join(gsSizes) %>%
  left_join(distinct(go, gs_name, shortest_path)) %>%
  dplyr::filter(
    FDR < 0.01, 
    shortest_path > 3, 
    NGenes < 1000,
    de_size >= 5
  ) %>%
  dplyr::select(
    `GO Gene Set` = gs_name,
    `Expressed Genes` = NGenes,
    `DE Genes` = de_size,
    Direction,
    PValue, FDR
    ) %>%
  pander(
    justify = "lrrlrr",
    caption = "Enriched GO Gene Sets, using the thresholds FDR < 0.01, at least 5 DE genes, a set size < 1000 and more than 3 steps back to the ontology root."
  )
```


### Exploration of Significant GO Gene Sets

```{r goFryUpset, fig.height=6, fig.width=10, fig.cap = "*UpSet plot indicating distribution of DE genes within significantly enriched terms from the GO gene sets. For this visualisation, GO terms were restricted to those with 15 or more DE genes, where this represented more than 5% of a gene set as being DE, along with an FDR < 0.02 and more than 3 steps back to the ontology root. The 20 largest GO terms satisfying these criteria are shown. The plot is  truncated at the right hand side for simplicity. A group of 28 genes is uniquely attributed to the Mitochondrial Envelop, with a further 18 being relatively unique to mRNA Metabolic Process. The next grouping of 15 genes are unique to Regulation Of Nucleobase-Containing Compound Metabolic Process followed by 25 genes, spread across two clusters of terms which largely represent Ribosomal activity. In between these are 13 genes uniquely associated with Anion Transport.*"}
goFry %>% 
  left_join(gsSizes) %>%
  left_join(distinct(go, gs_name, shortest_path)) %>%
  mutate(
    propDE = de_size / gs_size
  ) %>%
  dplyr::filter(
    de_size >= 15, 
    propDE > 0.05,
    FDR < 0.02,
    shortest_path > 3
  ) %>%
  dplyr::select(gs_name, de_id) %>%
  unnest(de_id) %>% 
  mutate(
    gs_name = str_remove(gs_name, "GO_"),
    gs_name = fct_lump(gs_name, n = 20)
    ) %>% 
  split(f = .$gs_name) %>% 
  lapply(magrittr::extract2, "de_id") %>% 
  fromList() %>% 
  upset(
    nsets = length(.), 
    nintersects = 20,
    order.by = "freq", 
    mb.ratio = c(0.6, 0.4),
    sets.x.label = "Number Of DE Genes"
    )
```

### Genes Associated with the Mitochondrial Envelope

```{r heatMito, fig.height=8, fig.width=8, fig.cap="*All DE genes associated with the GO term 'Mitochondrial Envelope'.*"}
go %>%
  dplyr::filter(
    gs_name == "GO_MITOCHONDRIAL_ENVELOPE",
    gene_id %in% dplyr::filter(deTable, DE)$gene_id
  ) %>%
  dplyr::select(
    gene_id, gene_name
  ) %>%
  left_join(
    cpmPostNorm %>%
      as.data.frame() %>%
      rownames_to_column("gene_id")
  ) %>%
  dplyr::select(gene_name, starts_with("Ps")) %>%
  as.data.frame() %>%
  column_to_rownames("gene_name") %>%
  as.matrix() %>%
  pheatmap(
    color = viridis_pal(option = "magma")(100),
    legend_breaks = c(seq(-2, 8, by = 2), max(.)),
    legend_labels = c(seq(-2, 8, by = 2), "logCPM\n"),
    labels_col = hmHeat %>%
      dplyr::select(starts_with("Ps")) %>% 
      colnames() %>% 
      enframe(name = NULL) %>% 
      dplyr::rename(sample = value) %>%
      left_join(dgeList$samples) %>%
      dplyr::select(sample, sampleID) %>%
      with(
        structure(sampleID, names = sample)
      ),
    annotation_col = dgeList$samples %>%
      dplyr::select(Genotype = genotype),
    cutree_rows = 4,
    cutree_cols = 2
  )
```


### Genes Associated with RNA Metabolic Process

```{r heatRnaMetab, fig.height=7, fig.width=8, fig.cap="*All DE genes associated with the GO term 'mRNA Metabolic Process'.*"}
go %>%
  dplyr::filter(
    gs_name == "GO_MRNA_METABOLIC_PROCESS",
    gene_id %in% dplyr::filter(deTable, DE)$gene_id
  ) %>%
  dplyr::select(
    gene_id, gene_name
  ) %>%
  left_join(
    cpmPostNorm %>%
      as.data.frame() %>%
      rownames_to_column("gene_id")
  ) %>%
  dplyr::select(gene_name, starts_with("Ps")) %>%
  as.data.frame() %>%
  column_to_rownames("gene_name") %>%
  as.matrix() %>%
  pheatmap(
    color = viridis_pal(option = "magma")(100),
    legend_breaks = c(seq(-2, 8, by = 2), max(.)),
    legend_labels = c(seq(-2, 8, by = 2), "logCPM\n"),
    labels_col = hmHeat %>%
      dplyr::select(starts_with("Ps")) %>% 
      colnames() %>% 
      enframe(name = NULL) %>% 
      dplyr::rename(sample = value) %>%
      left_join(dgeList$samples) %>%
      dplyr::select(sample, sampleID) %>%
      with(
        structure(sampleID, names = sample)
      ),
    annotation_col = dgeList$samples %>%
      dplyr::select(Genotype = genotype),
    cutree_rows = 4,
    cutree_cols = 2
  )
```

### Genes Associated with Regulation Of Nucleobase-Containing Compound Metabolic Process

```{r heatRegNucleo, fig.height=6, fig.width=8, fig.cap="*All DE genes associated with the GO term 'mRNA Metabolic Process'.*"}
go %>%
  dplyr::filter(
    gs_name == "GO_REGULATION_OF_NUCLEOBASE_CONTAINING_COMPOUND_METABOLIC_PROCESS",
    gene_id %in% dplyr::filter(deTable, DE)$gene_id
  ) %>%
  dplyr::select(
    gene_id, gene_name
  ) %>%
  left_join(
    cpmPostNorm %>%
      as.data.frame() %>%
      rownames_to_column("gene_id")
  ) %>%
  dplyr::select(gene_name, starts_with("Ps")) %>%
  as.data.frame() %>%
  column_to_rownames("gene_name") %>%
  as.matrix() %>%
  pheatmap(
    color = viridis_pal(option = "magma")(100),
    legend_breaks = c(seq(-2, 8, by = 2), max(.)),
    legend_labels = c(seq(-2, 8, by = 2), "logCPM\n"),
    labels_col = hmHeat %>%
      dplyr::select(starts_with("Ps")) %>% 
      colnames() %>% 
      enframe(name = NULL) %>% 
      dplyr::rename(sample = value) %>%
      left_join(dgeList$samples) %>%
      dplyr::select(sample, sampleID) %>%
      with(
        structure(sampleID, names = sample)
      ),
    annotation_col = dgeList$samples %>%
      dplyr::select(Genotype = genotype),
    cutree_rows = 4,
    cutree_cols = 2
  )
```

### Genes Associated with Anion Transport

```{r heatAnion, fig.height=7, fig.width=8, fig.cap="*All DE genes associated with the GO term 'Anion Transport'.*"}
go %>%
  dplyr::filter(
    gs_name == "GO_ANION_TRANSPORT",
    gene_id %in% dplyr::filter(deTable, DE)$gene_id
  ) %>%
  dplyr::select(
    gene_id, gene_name
  ) %>%
  left_join(
    cpmPostNorm %>%
      as.data.frame() %>%
      rownames_to_column("gene_id")
  ) %>%
  dplyr::select(gene_name, starts_with("Ps")) %>%
  as.data.frame() %>%
  column_to_rownames("gene_name") %>%
  as.matrix() %>%
  pheatmap(
    color = viridis_pal(option = "magma")(100),
    legend_breaks = c(seq(-2, 8, by = 2), max(.)),
    legend_labels = c(seq(-2, 8, by = 2), "logCPM\n"),
    labels_col = hmHeat %>%
      dplyr::select(starts_with("Ps")) %>% 
      colnames() %>% 
      enframe(name = NULL) %>% 
      dplyr::rename(sample = value) %>%
      left_join(dgeList$samples) %>%
      dplyr::select(sample, sampleID) %>%
      with(
        structure(sampleID, names = sample)
      ),
    annotation_col = dgeList$samples %>%
      dplyr::select(Genotype = genotype),
    cutree_rows = 3,
    cutree_cols = 2
  )
```


### Genes Associated with Ribosomal Activity

```{r heatRibo, fig.height=7, fig.width=8, fig.cap="*All DE genes associated with the GO term 'Cytosolic Ribosome'.*"}
go %>%
  dplyr::filter(
    gs_name == "GO_CYTOSOLIC_RIBOSOME",
    gene_id %in% dplyr::filter(deTable, DE)$gene_id
  ) %>%
  dplyr::select(
    gene_id, gene_name
  ) %>%
  left_join(
    cpmPostNorm %>%
      as.data.frame() %>%
      rownames_to_column("gene_id")
  ) %>%
  dplyr::select(gene_name, starts_with("Ps")) %>%
  as.data.frame() %>%
  column_to_rownames("gene_name") %>%
  as.matrix() %>%
  pheatmap(
    color = viridis_pal(option = "magma")(100),
    legend_breaks = c(seq(-2, 8, by = 2), max(.)),
    legend_labels = c(seq(-2, 8, by = 2), "logCPM\n"),
    labels_col = hmHeat %>%
      dplyr::select(starts_with("Ps")) %>% 
      colnames() %>% 
      enframe(name = NULL) %>% 
      dplyr::rename(sample = value) %>%
      left_join(dgeList$samples) %>%
      dplyr::select(sample, sampleID) %>%
      with(
        structure(sampleID, names = sample)
      ),
    annotation_col = dgeList$samples %>%
      dplyr::select(Genotype = genotype),
    cutree_rows = 3,
    cutree_cols = 2
  )
```


# Data Export

All enriched gene sets with an FDR adjusted p-value < 0.05 were exported as a single csv file.

```{r}
bind_rows(
  hmFry,
  kgFry,
  goFry
) %>%
  dplyr::filter(FDR < 0.05) %>%
  left_join(gsSizes) %>%
  dplyr::select(
    gs_name, NGenes, Direction, PValue, FDR,
    gene_symbol, de_size
  ) %>%
  mutate(
    DE = lapply(
      X = gene_symbol,
      FUN =  intersect, 
      y = dplyr::filter(deTable, DE)$gene_name
    ),
    DE = vapply(DE, paste, character(1), collapse = ";")
  ) %>%
  dplyr::select(
    `Gene Set` = gs_name, 
    `Nbr Detected Genes` = NGenes, 
    `Nbr DE Genes` = de_size, 
    Direction, PValue, FDR, 
    `DE Genes` = DE
  ) %>%
  write_csv(
    here::here("output", "Enrichment_Mutant_V_WT.csv")
  )
```