differential_expression.Rmd

---
title: "Differential expression across all donors"
author: "Davis J. McCarthy"
site: workflowr::wflow_site
---

Here, we will lok at differential expresion between clones across all donors at
the gene and gene set levels.

## Load libraries, data and DE results

```{r setup, include=FALSE, warning=FALSE, message=FALSE}
knitr::opts_chunk$set(echo = TRUE, warning = FALSE, message = FALSE)
library(tidyverse)
library(scater)
library(ggridges)
library(GenomicRanges)
library(RColorBrewer)
library(edgeR)
library(ggrepel)
library(ggcorrplot)
library(rlang)
library(limma)
library(org.Hs.eg.db)
library(ggforce)
library(superheat)
library(viridis)
library(IHW)
library(cowplot)
library(broom)
options(stringsAsFactors = FALSE)
```

Load the genewise differential expression results produced with the edgeR 
quasi-likelihood F test and gene set enrichment results produced with camera. 

```{r load-de}
params <- list()
params$callset <- "filt_lenient.cell_coverage_sites"
load(file.path("data/human_c6_v5p2.rdata"))
load(file.path("data/human_H_v5p2.rdata"))
load(file.path("data/human_c2_v5p2.rdata"))

de_res <- readRDS(paste0("data/de_analysis_FTv62/",
                         params$callset, 
                         ".de_results_unstimulated_cells.rds"))
```

Load SingleCellExpression objects with data used for differential expression 
analyses.

```{r load-sce}
fls <- list.files("data/sces")
fls <- fls[grepl(params$callset, fls)]
donors <- gsub(".*ce_([a-z]+)_.*", "\\1", fls)

sce_unst_list <- list()
for (don in donors) {
    sce_unst_list[[don]] <- readRDS(file.path("data/sces", 
        paste0("sce_", don, "_with_clone_assignments.", params$callset, ".rds")))
    cat(paste("reading", don, ":   ", ncol(sce_unst_list[[don]]), "cells.\n"))
}
```

The starting point for differential expression analysis was a set of 
`r length(donors)` donors, of which `r length(names(de_res$dge_list))` donors
had enough cells assigned to clones to conduct DE testing.

Summarise cell assignment information.

```{r assignments}
assignments_lst <- list()
for (don in donors) {
    assignments_lst[[don]] <- as_data_frame(
        colData(sce_unst_list[[don]])[, 
                                      c("donor_short_id", "highest_prob", 
                                        "assigned", "total_features",
                                        "total_counts_endogenous", "num_processed")])
}
assignments <- do.call("rbind", assignments_lst)
```

`r round(100 * mean(assignments$assigned != "unassigned"))`% of cells from these
donors are assigned with confidence to a clone.

Load donor info including evidence for selection dynamics in donors.

```{r load-donor-info}
df_donor_info <- read.table("data/donor_info_070818.txt")
```


## Genewise DE results

We first look at differential expression at the level of individual genes.

```{r genewise-de}
fdr_thresh <- 1
df_de_all_unst <- data_frame()
for (donor in names(de_res[["qlf_list"]])) {
    tmp <- de_res[["qlf_list"]][[donor]]$table
    tmp$gene <- rownames(de_res[["qlf_list"]][[donor]]$table)
    ihw_res <- ihw(PValue ~ logCPM, data = tmp, alpha = 0.05)
    tmp$FDR <- adj_pvalues(ihw_res)
    tmp <- tmp[tmp$FDR <= fdr_thresh,]
    if (nrow(tmp) > 0.5) {
        tmp[["donor"]] <- donor
        df_de_all_unst <- bind_rows(df_de_all_unst, tmp)
    }
}

df_ncells_de <- assignments %>% dplyr::filter(assigned != "unassigned", 
                              donor_short_id %in% names(de_res$qlf_list)) %>%
    group_by(donor_short_id) %>%
    summarise(n_cells = n())
colnames(df_ncells_de)[1] <- "donor"

fdr_thresh <- 0.1
df_de_sig_unst <- data_frame()
for (donor in names(de_res[["qlf_list"]])) {
    tmp <- de_res[["qlf_list"]][[donor]]$table
    tmp$gene <- rownames(de_res[["qlf_list"]][[donor]]$table)
    ihw_res <- ihw(PValue ~ logCPM, data = tmp, alpha = 0.05)
    tmp$FDR <- adj_pvalues(ihw_res)
    tmp <- tmp[tmp$FDR < fdr_thresh,]
    if (nrow(tmp) > 0.5) {
        tmp[["donor"]] <- donor
        df_de_sig_unst <- bind_rows(df_de_sig_unst, tmp)
    }
}

df_de_sig_unst %>% 
    group_by(gene) %>% 
    dplyr::mutate(id = paste0(donor, gene)) %>% distinct(id, .keep_all = TRUE) %>%
    summarise(n_donors = n()) %>% group_by(n_donors) %>%
    summarise(count = n()) %>%
ggplot(aes(x = n_donors, y = count)) +
    geom_segment(aes(x = n_donors, xend = n_donors, y = count, yend = 0.1), 
                 colour = "gray50") +
        geom_point(size = 3) +
    scale_y_log10(breaks = c(10, 100, 1000)) +
    scale_x_continuous(breaks = 0:11) +
    coord_cartesian(ylim = c(1, 2000)) +
    theme_classic(20) +
    xlab("Number of donors in which significant (FDR < 10%)")  +
    ylab("Number of genes") +
    ggtitle("edgeR QL F test DE results")

ggsave("figures/differential_expression/alldonors_de_n_sig_donors_n_sig_genes.png", 
       height = 7, width = 10)
ggsave("figures/differential_expression/alldonors_de_n_sig_donors_n_sig_genes.pdf", 
       height = 7, width = 10)
ggsave("figures/differential_expression/alldonors_de_n_sig_donors_n_sig_genes.svg", 
       height = 7, width = 10)
       
p1 <- df_de_sig_unst %>% 
    group_by(gene) %>% 
    dplyr::mutate(id = paste0(donor, gene)) %>% distinct(id, .keep_all = TRUE) %>%
    summarise(n_donors = n()) %>% group_by(n_donors) %>%
    summarise(count = n()) %>%
ggplot(aes(x = n_donors, y = count)) +
    geom_segment(aes(x = n_donors, xend = n_donors, y = count, yend = 0.1), 
                 colour = "gray50") +
        geom_point(size = 3) +
    scale_x_continuous(breaks = 0:11) +
    #coord_cartesian(ylim = c(1, 2200)) +
    theme_classic(16) +
    xlab("Number of donors significant (FDR < 10%)")  +
    ylab("Number of genes")
p1 

ggsave("figures/differential_expression/alldonors_de_n_sig_donors_n_sig_genes_linscale.png", 
       height = 7, width = 5.5)
ggsave("figures/differential_expression/alldonors_de_n_sig_donors_n_sig_genes_linscale.pdf", 
       height = 7, width = 5.5)
ggsave("figures/differential_expression/alldonors_de_n_sig_donors_n_sig_genes_linscale.svg", 
       height = 7, width = 5.5)
```


```{r, fig.height=14, fig.width=12}
p2 <- df_de_sig_unst %>% 
    group_by(gene) %>% 
    dplyr::mutate(id = paste0(donor, gene)) %>% distinct(id, .keep_all = TRUE) %>%
    summarise(n_donors = n()) %>% dplyr::arrange(gene, n_donors) %>% ungroup() %>%
    dplyr::mutate(gene = gsub("ENSG.*_", "", gene)) %>%
    dplyr::filter(n_donors > 7.5) %>%
    ggplot(aes(y = n_donors, x = reorder(gene, n_donors, max))) +
    geom_point(alpha = 0.7, size = 4) +
    scale_y_continuous(breaks = 7:11) +
    ggthemes::scale_colour_tableau() +
    coord_flip() +
    theme_bw(16) +
    xlab("Gene") + ylab("Number of donors significant")
p2

ggsave("figures/differential_expression/alldonors_de_n_sig_donors_topgenes.png", 
       height = 7, width = 5.5)
ggsave("figures/differential_expression/alldonors_de_n_sig_donors_topgenes.pdf", 
       height = 7, width = 5.5)
ggsave("figures/differential_expression/alldonors_de_n_sig_donors_topgenes.svg", 
       height = 7, width = 5.5)


#cowplot::plot_grid(p1, p2, rel_heights = c(0.4, 0.6))


df_donor_n_de <- df_de_sig_unst %>% 
    group_by(gene) %>% 
    dplyr::mutate(id = paste0(donor, gene)) %>% distinct(id, .keep_all = TRUE) %>%
    group_by(donor) %>%
    summarise(count = n())
no_de_donor <- unique(df_de_all_unst[["donor"]])[!(unique(df_de_all_unst[["donor"]]) %in% df_donor_n_de[["donor"]])]
df_donor_n_de <- rbind(df_donor_n_de, data_frame(donor = no_de_donor, count = 0))

```

Permute gene labels to get a null distribution.

```{r permute-de}
df_de_nsig <- df_de_all_unst %>% dplyr::filter(FDR < 0.1) %>% 
    dplyr::mutate(id = paste0(donor, gene)) %>% distinct(id, .keep_all = TRUE) %>%
    group_by(donor) %>% 
    summarise(n_sig = n())
df_nsig_ncells_de <- full_join(df_ncells_de, df_de_nsig)
df_nsig_ncells_de$n_sig[is.na(df_nsig_ncells_de$n_sig)] <- 0

permute_gene_labels <- function(gene_names, n_de)  {
    sampled_genes <- c()
    for (i in seq_along(n_de))
        sampled_genes <- c(sampled_genes, sample(gene_names, size = n_de[i]))
    tab <- table(table(sampled_genes))
    df <- data_frame(n_donors = 1:11, n_genes = 0)
    df[names(tab), 2] <- tab
    df
}


n_perm <- 1000
df_perm <- list()
for (i in seq_len(n_perm))
    df_perm[[i]] <- permute_gene_labels(rownames(de_res$qlf_list$vass$table), 
                                        df_nsig_ncells_de[["n_sig"]])
df_perm <- do.call("rbind", df_perm)
df_perm <- dplyr::mutate(df_perm, data_type = "permuted")

df_perm %>% group_by(n_donors) %>% summarise(min = min(n_genes), 
                                             median = median(n_genes),
                                             mean = mean(n_genes), 
                                             max = max(n_genes))

ppp <- df_de_sig_unst %>% 
    group_by(gene) %>% 
    dplyr::mutate(id = paste0(donor, gene)) %>% distinct(id, .keep_all = TRUE) %>%
    summarise(n_donors = n()) %>% group_by(n_donors) %>%
    summarise(n_genes = n()) %>% dplyr::mutate(data_type = "observed") %>%
ggplot(aes(x = n_donors, y = n_genes)) +
    # geom_segment(aes(x = n_donors, xend = n_donors, y = count, yend = 0.1), 
    #              colour = "gray50") +
#    geom_hline(yintercept = 0, linetype = 2) +
    geom_hline(yintercept = 0) +
    geom_boxplot(aes(group = n_donors, y = n_genes, colour = data_type), 
                 fill = "gray80", data = df_perm, show.legend = FALSE) +
    geom_point(aes(colour = data_type), shape = 17, size = 5) +
    scale_x_continuous(breaks = 0:11) +
    scale_y_sqrt(breaks = c(0, 10, 100, 500, 1000, 2000, 3000),
                 labels = c(0, 10, 100, 500, 1000, 2000, 3000),
                 limits = c(0, 4500)) +
    scale_colour_manual(name = '', 
                        values = c("observed" = "black", "permuted" = "gray50"), 
                        labels = c("observed", "permuted")) +
    coord_cartesian(ylim = c(0, 4500)) +
    theme_bw(18) +
    theme(legend.position = c(0.8, 0.88),
          panel.grid.major.x = element_blank(), 
          panel.grid.minor.x = element_blank()) +
    guides(colour = guide_legend(override.aes = list(shape = c(17, 19))),
           fill = FALSE, boxplot = FALSE) +
    xlab("Number of donors significant (FDR < 10%)")  +
    ylab("Number of genes")

ggsave("figures/differential_expression/alldonors_de_n_sig_donors_n_sig_genes_sqrtscale_perm.png", 
       height = 7, width = 10, plot = ppp)
ggsave("figures/differential_expression/alldonors_de_n_sig_donors_n_sig_genes_sqrtscale_perm.pdf", 
       height = 7, width = 10, plot = ppp)
ggsave("figures/differential_expression/alldonors_de_n_sig_donors_n_sig_genes_sqrtscale_perm.svg", 
       height = 7, width = 10, plot = ppp)

df_de_sig_unst %>% 
    group_by(gene) %>% 
    dplyr::mutate(id = paste0(donor, gene)) %>% distinct(id, .keep_all = TRUE) %>%
    summarise(n_donors = n()) %>% group_by(n_donors) %>%
    summarise(n_genes = n()) %>% dplyr::mutate(data_type = "observed") %>%
ggplot(aes(x = n_donors, y = n_genes)) +
    # geom_segment(aes(x = n_donors, xend = n_donors, y = count, yend = 0.1), 
    #              colour = "gray50") +
#    geom_hline(yintercept = 0, linetype = 2) +
    geom_hline(yintercept = 0) +
    geom_boxplot(aes(group = n_donors, y = n_genes, colour = data_type), 
                 fill = "gray80", data = df_perm, show.legend = FALSE) +
    geom_point(aes(colour = data_type), shape = 17, size = 5) +
    scale_x_continuous(breaks = 0:11) +
    scale_y_sqrt(breaks = c(0, 10, 100, 500, 1000, 2000, 3000),
                 labels = c(0, 10, 100, 500, 1000, 2000, 3000),
                 limits = c(0, 4500)) +
    scale_colour_manual(name = '', 
                        values = c("observed" = "black", "permuted" = "gray50"), 
                        labels = c("observed", "permuted")) +
    coord_cartesian(ylim = c(0, 4500)) +
    theme(legend.position = c(0.8, 0.88),
          panel.grid.major.x = element_blank(), 
          panel.grid.minor.x = element_blank()) +
    guides(colour = guide_legend(override.aes = list(shape = c(17, 19))),
           fill = FALSE, boxplot = FALSE) +
    xlab("Number of donors significant (FDR < 10%)")  +
    ylab("Number of genes")

ggsave("figures/differential_expression/alldonors_de_n_sig_donors_n_sig_genes_sqrtscale_perm_skinny.png", 
       height = 5.5, width = 6.5)

```

Look at recurrently DE genes.

```{r recurrent-de}
df_gene_n_de <- df_de_sig_unst %>% 
    group_by(gene) %>% 
    dplyr::mutate(id = paste0(donor, gene)) %>% distinct(id, .keep_all = TRUE) %>%
    group_by(gene) %>%
    summarise(count = n()) %>% 
    dplyr::arrange(desc(count)) %>%
    dplyr::mutate(ensembl_gene_id = gsub("_.*", "", gene),
                  hgnc_symbol = gsub(".*_", "", gene))
df_gene_n_de <- left_join(
    df_gene_n_de,
    dplyr::select(de_res$qlf_pairwise$joxm$clone2_clone1$table, 
                  ensembl_gene_id, hgnc_symbol, entrezid)
    )
df_gene_n_de <- dplyr::mutate(
    df_gene_n_de,
    cell_cycle_growth = (entrezid %in% 
                             c(Hs.H$HALLMARK_G2M_CHECKPOINT,
                               Hs.H$HALLMARK_MITOTIC_SPINDLE,
                               Hs.H$HALLMARK_E2F_TARGETS)),
    myc = (entrezid %in% c(Hs.H$HALLMARK_MYC_TARGETS_V1,
                           Hs.H$HALLMARK_MYC_TARGETS_V2)),
    emt = (entrezid %in% c(Hs.H$HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION))
)
df_gene_n_de %>% dplyr::filter(count >= 8) %>% 
  DT::datatable(.)
```

----------

## Camera results

First, aggregate gene set enrichment results across all donors.

```{r agg-camera-res}
fdr_thresh <- 1
df_camera_all_unst <- data_frame()
for (geneset in names(de_res[["camera"]])) {
    for (donor in names(de_res[["camera"]][[geneset]])) {
        for (coeff in names(de_res[["camera"]][[geneset]][[donor]])) {
            for (stat in names(de_res[["camera"]][[geneset]][[donor]][[coeff]])) {
                tmp <- de_res[["camera"]][[geneset]][[donor]][[coeff]][[stat]]
                tmp <- tmp[tmp$FDR <= fdr_thresh,]
                if (nrow(tmp) > 0.5) {
                    tmp[["collection"]] <- geneset
                    tmp[["geneset"]] <- rownames(tmp)
                    tmp[["coeff"]] <- coeff
                    tmp[["donor"]] <- donor
                    tmp[["stat"]] <- stat
                    df_camera_all_unst <- bind_rows(df_camera_all_unst, tmp)
                }
            }
        }
    }
}

fdr_thresh <- 0.05
df_camera_sig_unst <- data_frame()
for (geneset in names(de_res[["camera"]])) {
    for (donor in names(de_res[["camera"]][[geneset]])) {
        for (coeff in names(de_res[["camera"]][[geneset]][[donor]])) {
            for (stat in names(de_res[["camera"]][[geneset]][[donor]][[coeff]])) {
                tmp <- de_res[["camera"]][[geneset]][[donor]][[coeff]][[stat]]
                tmp <- tmp[tmp$FDR <= fdr_thresh,]
                if (nrow(tmp) > 0.5) {
                    tmp[["collection"]] <- geneset
                    tmp[["geneset"]] <- rownames(tmp)
                    tmp[["coeff"]] <- coeff
                    tmp[["donor"]] <- donor
                    tmp[["stat"]] <- stat
                    df_camera_sig_unst <- bind_rows(df_camera_sig_unst, tmp)
                }
            }
        }
    }
}

df_camera_sig_unst <- dplyr::mutate(
    df_camera_sig_unst,
    contrast = gsub("_", " - ", coeff),
    msigdb_collection = plyr::mapvalues(collection, from = c("c2", "c6", "H"), to = c("MSigDB curated (c2)", "MSigDB oncogenic (c6)", "MSigDB Hallmark")))

df_camera_all_unst <- dplyr::mutate(
    df_camera_all_unst,
    contrast = gsub("_", " - ", coeff),
    msigdb_collection = plyr::mapvalues(collection, from = c("c2", "c6", "H"), to = c("MSigDB curated (c2)", "MSigDB oncogenic (c6)", "MSigDB Hallmark")))
```

We now have a dataframe for significant (FDR <5%) results from the camera 
gene set enrichment results.

```{r}
head(df_camera_sig_unst)
```

And a dataframe with all results.

```{r}
head(df_camera_all_unst)
```

For now, focus on gene set enrichment results computed using log-fold change 
statistics for pairwise comparisons of clones estimated from the edgeR QL-F 
models.

We can look at all significant results summarised by donor, geneset and pairwise
contrast of clones.

```{r camera-allcontr-allsets-sig-bydonor-pvals, fig.height=17, fig.width=11}
df_camera_sig_unst %>% 
    dplyr::filter(stat == "logFC") %>%
    dplyr::mutate(donor = factor(donor, levels = rev(levels(factor(donor))))) %>%
ggplot(aes(y = -log10(PValue), x = donor, colour = contrast)) +
    geom_sina(alpha = 0.7) +
    facet_grid(contrast ~ msigdb_collection) + 
    scale_colour_brewer(palette = "Accent") +
    coord_flip() + theme_bw()
```

Similarly, we can look at all results summarised by donor, geneset and pairwise
contrast of clones.

```{r camera-allcontr-allsets-all-bydonor-fdr, fig.height=17, fig.width=11}
df_camera_all_unst %>%
    dplyr::filter(stat == "logFC") %>%
    dplyr::mutate(donor = factor(donor, levels = rev(levels(factor(donor))))) %>%
ggplot(aes(y = -log10(FDR), x = donor, colour = contrast)) +
    geom_sina(alpha = 0.7) +
    geom_hline(yintercept = -log10(0.05), linetype = 2, colour = "firebrick") +
    facet_grid(contrast ~ msigdb_collection, scales = "free_x") + 
    scale_colour_brewer(palette = "Accent") +
    coord_flip() + theme_bw()

ggsave("figures/differential_expression/alldonors_camera_enrichment_by_donor_all_results.png", 
       height = 16, width = 14)
ggsave("figures/differential_expression/alldonors_camera_enrichment_by_donor_all_results.pdf", 
       height = 16, width = 14)
ggsave("figures/differential_expression/alldonors_camera_enrichment_by_donor_all_results.svg", 
       height = 16, width = 14)
```

We can check the number of significant gene sets for each donor, for each MSigDB
gene set collection.

```{r camera-nsig-donor}
df_camera_sig_unst %>% 
    dplyr::filter(stat == "logFC") %>%
    dplyr::filter(FDR < 0.05) %>%
    group_by(donor, msigdb_collection) %>%
    summarise(n_sig = n()) %>% print(n = Inf)
```

We can look at the number of significant gene sets for each donor.

```{r alldonors_camera_enrichment_by_donor_simple, fig.height=12, fig.width=8}
## simpler version
df_camera_all_unst %>% 
    dplyr::filter(stat == "logFC") %>%
    group_by(donor, msigdb_collection) %>%
    summarise(n_sig = sum(FDR < 0.05)) %>% ungroup() %>% 
    dplyr::mutate(donor = factor(donor, levels = rev(levels(factor(donor))))) %>%
    ggplot(aes(y = n_sig, x = donor)) +
    geom_point(alpha = 1, size = 4) +
    facet_wrap(~ msigdb_collection, scales = "free_x") + 
    scale_fill_brewer(palette = "Accent") +
    coord_flip() +
    theme_bw(16) +
    xlab("Donor") + ylab("Number of significant genesets (FDR < 5%)") +
    ggtitle("Camera MSigDB gene set enrichment by donor")

ggsave("figures/differential_expression/alldonors_camera_enrichment_by_donor_simple.png", 
       height = 7, width = 10)
ggsave("figures/differential_expression/alldonors_camera_enrichment_by_donor_simple.pdf", 
       height = 7, width = 10)
ggsave("figures/differential_expression/alldonors_camera_enrichment_by_donor_simple.svg", 
       height = 7, width = 10)
```

We can look at the effect of the the number of cells for each donor on the 
DE results obtained.

```{r alldonors_camera_enrichment_by_donor_simple_size_by_ncells, fig.height=12, fig.width=8}
ncells_by_donor <- rep(NA, length(sce_unst_list))
names(ncells_by_donor) <- names(sce_unst_list)
for (don in names(sce_unst_list))
    ncells_by_donor[don] <- ncol(sce_unst_list[[don]])

df_camera_all_unst %>% 
    dplyr::filter(stat == "logFC") %>%
    group_by(donor, msigdb_collection) %>%
    summarise(n_sig = sum(FDR < 0.05)) %>% ungroup() -> df_to_plot
df_to_plot <- inner_join(df_to_plot, 
                        data_frame(donor = names(ncells_by_donor),
                                   ncells = ncells_by_donor))
df_to_plot %>%
    dplyr::mutate(donor = factor(donor, levels = rev(levels(factor(donor))))) %>%
    ggplot(aes(y = n_sig, x = donor, size = ncells)) +
    geom_point(alpha = 1) +
    facet_wrap(~ msigdb_collection, scales = "free_x") + 
    scale_fill_brewer(palette = "Accent") +
    coord_flip() +
    theme_bw(16) +
    xlab("Donor") + ylab("Number of significant genesets (FDR < 5%)") +
    ggtitle("Camera MSigDB gene set enrichment by donor")

ggsave("figures/differential_expression/alldonors_camera_enrichment_by_donor_simple_size_by_ncells.png", 
       height = 7, width = 10)
ggsave("figures/differential_expression/alldonors_camera_enrichment_by_donor_simple_size_by_ncells.pdf", 
       height = 7, width = 10)
ggsave("figures/differential_expression/alldonors_camera_enrichment_by_donor_simple_size_by_ncells.svg", 
       height = 7, width = 10)

```

### Hallmark gene set

Focus now on looking at DE results for the MSigDB Hallmark gene set (50 of the 
best-characterised gene sets as determined by MSigDB).

Look at the gene sets that are found to be enriched in multiple donors.

```{r alldonors_camera_enrichment_H_by_geneset, fig.height=7, fig.width=10}

## Hallmark geneset
df_camera_sig_unst %>% dplyr::filter(collection == "H") %>% 
    dplyr::filter(stat == "logFC") %>%
    group_by(geneset) %>% 
    dplyr::mutate(id = paste0(donor, geneset)) %>% distinct(id, .keep_all = TRUE) %>%
    summarise(n_donors = n()) %>% dplyr::arrange(geneset, n_donors) %>% ungroup() %>%
    dplyr::mutate(geneset = gsub("_", " ", gsub("HALLMARK_", "", geneset))) %>%
ggplot(aes(y = n_donors, x = reorder(geneset, n_donors, max))) +
    geom_point(alpha = 0.7, size = 4) +
    ggthemes::scale_colour_tableau() +
    coord_flip() +
    theme_bw(14) +
    xlab("Gene set") + ylab("Number of donors significant")

ggsave("figures/differential_expression/alldonors_camera_enrichment_H_by_geneset.png", 
       height = 7, width = 9.5)
ggsave("figures/differential_expression/alldonors_camera_enrichment_H_by_geneset.pdf", 
       height = 7, width = 9.5)
ggsave("figures/differential_expression/alldonors_camera_enrichment_H_by_geneset.svg", 
       height = 7, width = 9.5)
## number of donors with at least one significant geneset
tmp <- df_camera_sig_unst %>% dplyr::filter(collection == "H") %>% 
    dplyr::filter(stat == "logFC") %>%
    group_by(geneset)
unique(tmp[["donor"]])
```

`r length(unique(tmp[["donor"]]))` donors have at least one significantly 
enriched Hallmark gene set.

For gene sets related directly to cell cycle and growth, we see contrasts being
both up- and down- regulated, but for EMT, coagulation and angiogenesis pathways,
we only see these down-regulated.

```{r alldonors_camera_enrichment_H_by_geneset_by_dir, fig.height=10, fig.width=16}
df_camera_sig_unst %>% dplyr::filter(collection == "H") %>% 
  dplyr::filter(stat == "logFC") %>%
  group_by(geneset, Direction) %>% 
  dplyr::mutate(id = paste0(donor, geneset)) %>% 
  summarise(n_donors = n()) %>% dplyr::arrange(geneset, n_donors) %>% ungroup() %>%
  ggplot(aes(y = n_donors, x = reorder(geneset, n_donors, max),
             colour = Direction)) +
  geom_point(alpha = 0.7, size = 4, position = position_dodge(width = 0.5)) +
  ggthemes::scale_colour_tableau() +
  coord_flip() +
  theme_bw(16) +
  xlab("Gene set") + ylab("Number of donors significant")  +
  ggtitle("Camera MSigDB Hallmark gene set enrichment")
```



## Heatmap of results for camera Hallmark geneset testing

We can get an overview of all the Hallmark gene set results by producing a 
heatmap, first showing just the significant (FDR < 5%) results across all 
donors and pairwise contrasts of clones.

```{r top_genesets_H_direction_heatmap, fig.height=7, fig.width=12}
repeated_sig_H_genesets <- df_camera_sig_unst %>% 
    dplyr::filter(collection == "H", stat == "logFC") %>% 
    group_by(geneset) %>% 
    dplyr::mutate(id = paste0(donor, geneset)) %>% distinct(id, .keep_all = TRUE) %>%
    summarise(n_donors = n()) %>% dplyr::arrange(n_donors) %>% 
    dplyr::filter(n_donors > 0.5) 
repeated_sig_H_genesets_vec <- unique(repeated_sig_H_genesets[["geneset"]])
repeated_sig_H_genesets_vec <- gsub("_", " ", gsub("HALLMARK_", "", 
                                                   repeated_sig_H_genesets_vec))
     
df_camera_sig_unst %>% 
    dplyr::mutate(geneset = gsub("_", " ", gsub("HALLMARK_", "", geneset))) %>%
    dplyr::mutate(geneset = 
                      factor(geneset, levels = repeated_sig_H_genesets_vec)) %>%
    dplyr::filter(geneset %in% repeated_sig_H_genesets_vec) %>%
    dplyr::mutate(id = paste0(donor, ": ", contrast)) ->
    df_4_heatmap

div_lines <- gsub(": c.*", "",
     sort(unique(paste0(df_4_heatmap[["donor"]], ": ", 
                        df_4_heatmap[["contrast"]])))) %>% table %>% cumsum + 0.5
    
df_4_heatmap %>%
    ggplot(aes(x = id, y = geneset, fill = Direction)) +
    geom_tile() +
    geom_vline(xintercept = div_lines, colour = "gray70") +
    scale_fill_manual(values = c("lightgoldenrod1", "sienna1")) +
    theme(axis.text.x = element_text(angle = 60, hjust = 1))

ggsave("figures/differential_expression/top_genesets_H_direction_heatmap.png", height = 6, width = 12)
ggsave("figures/differential_expression/top_genesets_H_direction_heatmap.pdf", height = 6, width = 12)
ggsave("figures/differential_expression/top_genesets_H_direction_heatmap.svg", height = 6, width = 12)
```

We can do the same for all results for the gene sets that are significantly 
enriched in at least two donors.

```{r top_genesets_H_direction_heatmap_all_contrasts, fig.height=9, fig.width=18}
df_camera_all_unst %>% 
    dplyr::mutate(geneset = gsub("_", " ", gsub("HALLMARK_", "", geneset))) %>%
    dplyr::mutate(geneset = 
                      factor(geneset, levels = repeated_sig_H_genesets_vec)) %>%
    dplyr::filter(geneset %in% repeated_sig_H_genesets_vec) %>%
    dplyr::mutate(id = paste0(donor, ": ", contrast)) ->
    df_4_heatmap_all

df_4_heatmap_all <- dplyr::mutate(
    df_4_heatmap_all,
    minlog10P = cut(-log10(PValue), breaks = c(0, 1, 2, 3, 4, 5, 30)))

div_lines_all <- gsub(": c.*", "",
     sort(unique(paste0(df_4_heatmap_all[["donor"]], ": ", 
                        df_4_heatmap_all[["contrast"]])))) %>% table %>% cumsum + 0.5

pp <- df_4_heatmap_all %>%
    ggplot(aes(x = id, y = geneset, fill = Direction, alpha = minlog10P)) +
    geom_tile() +
    geom_point(alpha = 1, data = df_4_heatmap, pch = 19, size = 0.5, show.legend = FALSE) +
    geom_vline(xintercept = div_lines_all, colour = "gray70") +
    scale_fill_manual(values = c("lightgoldenrod1", "sienna1")) +
    scale_alpha_discrete(name = "-log10(P)") +
    ylab("Gene set") +
    xlab("Donor and clone comparison") +
    theme(axis.text.x = element_text(angle = 60, hjust = 1),
          legend.position = "right")
pp    

ggsave("figures/differential_expression/top_genesets_H_direction_heatmap_all_contrasts.png", height = 9, width = 20)
ggsave("figures/differential_expression/top_genesets_H_direction_heatmap_all_contrasts.pdf", height = 9, width = 20)
ggsave("figures/differential_expression/top_genesets_H_direction_heatmap_all_contrasts.svg", height = 9, width = 20)
```

Finally, we can add a panel to this figure showing the number of donors in which
each of these gene sets is significantly enriched.

```{r top_genesets_H_direction_heatmap_all_contrasts_with_nsig_donors, fig.height=9, fig.width=20}
## Hallmark geneset
pp_nsig <- df_camera_sig_unst %>% dplyr::filter(collection == "H") %>% 
    dplyr::filter(stat == "logFC") %>%
    group_by(geneset) %>% 
    dplyr::mutate(id = paste0(donor, geneset)) %>% distinct(id, .keep_all = TRUE) %>%
    summarise(n_donors = n()) %>% dplyr::arrange(geneset, n_donors) %>% ungroup() %>%
    dplyr::mutate(geneset = gsub("_", " ", gsub("HALLMARK_", "", geneset))) %>%
    ggplot(aes(y = n_donors, x = reorder(geneset, n_donors, max))) +
    geom_hline(yintercept = 0, colour = "gray50") +
    geom_segment(aes(xend = reorder(geneset, n_donors, max), yend = 0),
                 colour = "gray50") +
    geom_point(size = 4, colour = "gray30", alpha = 1) +
    ggthemes::scale_colour_tableau() +
    coord_flip() +
    xlab("Gene set") + ylab("Number of donors significant") +
    theme(axis.title.y = element_blank(),
          axis.text.y = element_blank(),
          axis.ticks.y = element_blank(),
          axis.line.y = element_blank())

prow <- plot_grid(pp + theme(legend.position = "none"), 
                  pp_nsig, align = 'h', rel_widths = c(7, 1))
lgnd <- get_legend(pp)
plot_grid(prow, lgnd, rel_widths = c(3, .3))

ggsave("figures/differential_expression/top_genesets_H_direction_heatmap_all_contrasts_with_nsig_donors.png", height = 9, width = 20)
ggsave("figures/differential_expression/top_genesets_H_direction_heatmap_all_contrasts_with_nsig_donors.pdf", height = 9, width = 20)
ggsave("figures/differential_expression/top_genesets_H_direction_heatmap_all_contrasts_with_nsig_donors.svg", height = 9, width = 20)
```


## Correlation of gene set results and genes contained

Let's look at the correlation between gene set results (Spearman correlation of
signed -log10(P-values) from _camera_ tests) and compare to the proportion of 
genes overlapping between pairs of gene sets.

```{r corr-maps, fig.height=9, fig.width=13}
repeated_sig_H_genesets_vec2 <- paste0("HALLMARK_", 
                                       gsub(" ", "_", repeated_sig_H_genesets_vec))
## all results
df_H_pvals <- df_camera_all_unst %>% dplyr::filter(collection == "H") %>% 
  dplyr::filter(stat == "logFC", geneset %in% repeated_sig_H_genesets_vec2) %>%
  dplyr::mutate(donor_coeff = paste(donor, coeff, sep = "."),
                sign = ifelse(Direction == "Down", -1, 1),
                signed_P = sign * -log10(PValue)) %>%
  dplyr::select(geneset, donor_coeff, signed_P) %>%
  tidyr::spread(key = donor_coeff, value = signed_P)

mat_H_pvals <- as.matrix(df_H_pvals[, -1])
rownames(mat_H_pvals) <- gsub("_", " ", gsub("HALLMARK_", "", df_H_pvals[[1]]))
cor_H_pvals <- cor(t(mat_H_pvals), method = "spearman")
p.mat <- cor_pmat(t(mat_H_pvals))
ggcorrplot(cor_H_pvals, hc.order = TRUE, p.mat = p.mat, insig = "blank") +
  theme(panel.grid.major = element_blank(),
        panel.grid.minor = element_blank())

hclust_cor <- hclust(as.dist(1 - cor_H_pvals))
corrplot1 <- ggcorrplot(cor_H_pvals[hclust_cor$order, hclust_cor$order], 
           p.mat = p.mat[hclust_cor$order, hclust_cor$order], insig = "blank") +
  theme(panel.grid.major = element_blank(),
        panel.grid.minor = element_blank())

mat_H_gene_overlap <- matrix(nrow = nrow(cor_H_pvals), ncol = ncol(cor_H_pvals),
                             dimnames = dimnames(cor_H_pvals))
for (i in seq_along(repeated_sig_H_genesets_vec2)) {
  for (j in seq_along(repeated_sig_H_genesets_vec2)) {
    gs1 <- paste0("HALLMARK_", gsub(" ", "_", rownames(mat_H_gene_overlap)[i]))
    gs2 <- paste0("HALLMARK_", gsub(" ", "_", rownames(mat_H_gene_overlap)[j]))
    mat_H_gene_overlap[i, j] <- mean(Hs.H[[gs1]] %in% Hs.H[[gs2]])
  }
}

corrplot2 <- ggcorrplot(mat_H_gene_overlap[hclust_cor$order, hclust_cor$order]) +
  scale_fill_gradient(name = "Gene set\noverlap", low = "white", high = "black") +
  theme(panel.grid.major = element_blank(),
        panel.grid.minor = element_blank())
corrplot2
corrplot3 <- corrplot2 + theme(axis.text.y = element_blank())
```

```{r corr-plot-combined, fig.height=9, fig.width=20}
plot_grid(corrplot1 + theme(plot.margin = unit(c(0,0,0,0), "cm")), 
          corrplot3 + theme(plot.margin = unit(c(0,0,0,0), "cm")), 
          align = "h", axis = "b", rel_widths = c(0.58, 0.42))

ggsave("figures/differential_expression/top_genesets_H_corrplots.png", height = 9, width = 20)
ggsave("figures/differential_expression/top_genesets_H_corrplots.pdf", height = 9, width = 20)
ggsave("figures/differential_expression/top_genesets_H_corrplots.svg", height = 9, width = 20)
```

Plot gene set correlation with the number of donors in which each gene set is 
significant.

```{r, fig.height=9, fig.width=11}
pp_nsig <- df_camera_sig_unst %>% dplyr::filter(collection == "H") %>% 
    dplyr::filter(stat == "logFC") %>%
    group_by(geneset) %>% 
    dplyr::mutate(id = paste0(donor, geneset)) %>% distinct(id, .keep_all = TRUE) %>%
    summarise(n_donors = n()) %>% dplyr::arrange(geneset, n_donors) %>% ungroup() %>%
    dplyr::mutate(geneset = gsub("_", " ", gsub("HALLMARK_", "", geneset))) %>%
    dplyr::mutate(geneset = factor(
      geneset, levels = rownames(mat_H_gene_overlap)[hclust_cor$order])) %>%
    ggplot(aes(y = n_donors, x = geneset)) +
    geom_hline(yintercept = 0, colour = "gray50") +
    geom_segment(aes(xend = geneset, yend = 0),
                 colour = "gray50") +
    geom_point(size = 4, colour = "gray30", alpha = 1) +
    ggthemes::scale_colour_tableau() +
    coord_flip() +
    xlab("Gene set") + ylab("Number of donors\nsignificant") +
    theme(axis.title.y = element_blank(),
          axis.text.y = element_blank(),
          axis.ticks.y = element_blank(),
          axis.line.y = element_blank())

ggdraw() +
  draw_plot(corrplot1 + theme(legend.position = "top"), 
            x = 0,  y = 0, width = 0.8, scale = 1) +
  draw_plot(pp_nsig, 
            x = 0.785,  y = 0.213, width = 0.2, height = 0.685)

ggsave("figures/differential_expression/top_genesets_H_corrplot_with_nsig_donor.png", 
       height = 9, width = 11)
ggsave("figures/differential_expression/top_genesets_H_corrplot_with_nsig_donor.pdf", 
       height = 9, width = 11)

```



## Linking DE to selection

```{r}
df_donor_info <- read.table("data/donor_info_070818.txt")
df_donor_info <- as_data_frame(df_donor_info)
df_donor_info$donor <- df_donor_info$donor_short

df_ncells_de <- assignments %>% dplyr::filter(assigned != "unassigned", 
                              donor_short_id %in% names(de_res$qlf_list)) %>%
    group_by(donor_short_id) %>%
    summarise(n_cells = n())
colnames(df_ncells_de)[1] <- "donor"

df_prop_assigned <- assignments %>% 
  dplyr::filter(donor_short_id %in% names(de_res$qlf_list)) %>%
    group_by(donor_short_id) %>%
    summarise(prop_assigned = mean(assigned != "unassigned"))
colnames(df_prop_assigned)[1] <- "donor"

df_nvars_by_cat <- readr::read_tsv("output/nvars_by_category_by_donor.tsv")
df_nvars_by_cat_wd <- tidyr::spread(
  df_nvars_by_cat[, 1:3], consequence, n_vars_all_genes)
df_nvars_by_cat_wd <- left_join(
  summarise(group_by(df_nvars_by_cat, donor), nvars_all = sum(n_vars_all_genes)),
  df_nvars_by_cat_wd
)
df_nvars_by_cat_wd <- df_nvars_by_cat %>% 
  dplyr::filter(consequence %in% c("missense", "splicing", "nonsense")) %>%
  group_by(donor) %>%
  summarise(nvars_misnonspli = sum(n_vars_all_genes)) %>%
  left_join(., df_nvars_by_cat_wd)

df_donor_info <- left_join(df_ncells_de, df_donor_info)
df_donor_info <- left_join(df_prop_assigned, df_donor_info)
df_donor_info <- left_join(df_donor_n_de, df_donor_info)
df_donor_info$n_de_genes <- df_donor_info$count
df_donor_info <- left_join(df_donor_info, df_nvars_by_cat_wd)

nbglm_nde <- MASS::glm.nb(n_de_genes ~ n_cells, data = df_donor_info)
df_nbglm_nde <- broom::augment(nbglm_nde) %>%
  left_join(df_donor_info)

## n_de vs n_cells
df_nbglm_nde %>%
  dplyr::mutate(selection = factor(
    selection, levels = c("neutral", "undetermined", "selected"))) %>%
ggplot(aes(x = n_cells, y = n_de_genes, fill = selection)) +
  geom_smooth(aes(group = 1), colour = "firebrick", method = "lm", level = 0.9) +
  geom_point(size = 3, shape = 21) +
  ylab("Number of DE genes") +
  xlab("Number of cells") +
  scale_fill_manual(values = c("dodgerblue", "#CCCCCC", "dodgerblue4"))

ggsave("figures/differential_expression/n_de_genes_vs_n_cells.png", 
       height = 5.5, width = 5.5)

## selection, n_de resid boxplot
df_nbglm_nde %>%
  dplyr::mutate(selection = factor(
    selection, levels = c("neutral", "undetermined", "selected"))) %>%
ggplot(aes(x = selection, y = .resid)) +
  geom_violin(aes(fill = selection), alpha = 0.7) +
  geom_boxplot(outlier.alpha = 0, width = 0.2) +
  ggbeeswarm::geom_quasirandom(aes(fill = selection), size = 3, shape = 21) +
  ylab("Number of DE genes (residual from NB GLM)") +
  xlab("Inferred selection status") +
  scale_fill_manual(values = c("dodgerblue", "#CCCCCC", "dodgerblue4")) +
  coord_flip()

ggsave("figures/differential_expression/n_de_resid_selection_boxplot.png", 
       height = 4.5, width = 6.5)

summary(lm(.resid ~ selection, data = df_nbglm_nde))

## selection, n_de (sqrt scale) boxplot
df_nbglm_nde %>%
  dplyr::mutate(selection = factor(
    selection, levels = c("neutral", "undetermined", "selected"))) %>%
ggplot(aes(x = selection, y = n_de_genes)) +
  geom_violin(aes(fill = selection), alpha = 0.7) +
  geom_boxplot(outlier.alpha = 0, width = 0.2) +
  ggbeeswarm::geom_quasirandom(aes(fill = selection), size = 3, shape = 21) +
  ylab("Number of DE genes") +
  xlab("Inferred selection status") +
  scale_y_sqrt(breaks = c(0, 100, 500, 1000, 1500, 2000, 2500)) +
  scale_fill_manual(values = c("dodgerblue", "#CCCCCC", "dodgerblue4")) +
  coord_flip()

ggsave("figures/differential_expression/n_de_sqrt_selection_boxplot.png", 
       height = 5.5, width = 6.5)

## n_de (resids) vs goodness of fit cumul. mutation model
df_nbglm_nde %>%
  dplyr::mutate(selection = factor(
    selection, levels = c("neutral", "undetermined", "selected"))) %>%
ggplot(aes(x = rsq_ntrtestr, y = .resid, fill = selection)) +
  geom_smooth(aes(group = 1), colour = "firebrick", method = "lm", level = 0.9) +
  geom_point(size = 3, shape = 21) +
  ylab("Number of DE genes (residual from NB GLM)") +
  xlab("Goodness of fit: cumulative mutations") +
  scale_fill_manual(values = c("dodgerblue", "#CCCCCC", "dodgerblue4"))

ggsave("figures/differential_expression/n_de_resid_selection_vs_gof_cumul_mut_model.png", 
       height = 6.5, width = 5.5)


## n_de (sqrt scale) vs goodness of fit cumul. mutation model
df_nbglm_nde %>%
  dplyr::mutate(selection = factor(
    selection, levels = c("neutral", "undetermined", "selected"))) %>%
ggplot(aes(x = rsq_ntrtestr, y = n_de_genes, fill = selection)) +
  geom_smooth(aes(group = 1), colour = "firebrick", method = "lm", level = 0.9) +
  geom_point(size = 3, shape = 21) +
  ylab("Number of DE genes") +
  xlab("Goodness of fit: cumulative mutations") +
  scale_fill_manual(values = c("dodgerblue", "#CCCCCC", "dodgerblue4"))

ggsave("figures/differential_expression/n_de_sqrt_selection_vs_gof_cumul_mut_model.png", 
       height = 5.5, width = 6.5)

## n_de (resids) vs goodness of fit NB model
df_nbglm_nde %>%
  dplyr::mutate(selection = factor(
    selection, levels = c("neutral", "undetermined", "selected"))) %>%
ggplot(aes(x = rsq_negbinfit, y = .resid, fill = selection)) +
  geom_smooth(aes(group = 1), colour = "firebrick", method = "lm", level = 0.9) +
  geom_point(size = 3, shape = 21) +
  ylab("Number of DE genes (residual from NB GLM)") +
  xlab("Goodness of fit: negative binomial distribution") +
  scale_fill_manual(values = c("dodgerblue", "#CCCCCC", "dodgerblue4"))

ggsave("figures/differential_expression/n_de_resid_selection_vs_gof_negbin_model.png", 
       height = 5.5, width = 6.5)

## n_de (resids) vs mutational load (all)
df_nbglm_nde %>%
  dplyr::mutate(selection = factor(
    selection, levels = c("neutral", "undetermined", "selected"))) %>%
ggplot(aes(x = nvars_all, y = .resid, fill = selection)) +
  geom_smooth(aes(group = 1), colour = "firebrick", method = "lm", level = 0.9) +
  geom_point(size = 3, shape = 21) +
  ylab("Number of DE genes (residual from NB GLM)") +
  xlab("Number of somatic variants") +
  scale_fill_manual(values = c("dodgerblue", "#CCCCCC", "dodgerblue4")) +
  scale_x_log10(breaks = c(5, 10, 20, 50, 100, 500))

ggsave("figures/differential_expression/n_de_resid_selection_vs_n_somatic_vars_all.png", 
       height = 5.5, width = 6.5)

## n_de (sqrt scale) vs mutational load (all)
df_nbglm_nde %>%
  dplyr::mutate(selection = factor(
    selection, levels = c("neutral", "undetermined", "selected"))) %>%
ggplot(aes(x = nvars_all, y = n_de_genes, fill = selection)) +
  geom_smooth(aes(group = 1), colour = "firebrick", method = "lm", level = 0.9) +
  geom_point(size = 3, shape = 21) +
  ylab("Number of DE genes") +
  xlab("Number of somatic variants") +
  scale_fill_manual(values = c("dodgerblue", "#CCCCCC", "dodgerblue4")) +
  scale_x_log10(breaks = c(5, 10, 20, 50, 100, 500)) +
  scale_y_sqrt(breaks = c(10, 100, 500, 1000, 1500, 2000, 2500))

ggsave("figures/differential_expression/n_de_sqrt_selection_vs_n_somatic_vars_all.png", 
       height = 5.5, width = 6.5)

## n_de (resids) vs mutational load (missense)
df_nbglm_nde %>%
  dplyr::mutate(selection = factor(
    selection, levels = c("neutral", "undetermined", "selected"))) %>%
ggplot(aes(x = missense, y = .resid, fill = selection)) +
  geom_smooth(aes(group = 1), colour = "firebrick", method = "lm", level = 0.9) +
  geom_point(size = 3, shape = 21) +
  ylab("Number of DE genes (residual from NB GLM)") +
  xlab("Number of somatic missense variants") +
  scale_fill_manual(values = c("dodgerblue", "#CCCCCC", "dodgerblue4")) +
  scale_x_log10(breaks = c(5, 10, 20, 50, 100, 500))

ggsave("figures/differential_expression/n_de_resid_selection_vs_n_somatic_vars_missense.png", 
       height = 5.5, width = 6.5)

## n_de (resids) vs mutational load (missense, nonsense, splicing)
df_nbglm_nde %>%
  dplyr::mutate(selection = factor(
    selection, levels = c("neutral", "undetermined", "selected"))) %>%
ggplot(aes(x = nvars_misnonspli, y = .resid, fill = selection)) +
  geom_smooth(aes(group = 1), colour = "firebrick", method = "lm", level = 0.9) +
  geom_point(size = 3, shape = 21) +
  ylab("Number of DE genes (residual from NB GLM)") +
  xlab("Number of missense, nonsense & splicing variants") +
  scale_fill_manual(values = c("dodgerblue", "#CCCCCC", "dodgerblue4")) +
  scale_x_log10(breaks = c(5, 10, 20, 50, 100, 500))

ggsave("figures/differential_expression/n_de_resid_selection_vs_n_somatic_vars_misnonspli.png", 
       height = 6.5, width = 5.5)

# 
# fdr_thresh <- 0.1
# df_de_sig_unst <- data_frame()
# for (donor in names(de_res[["qlf_list"]])) {
#     tmp <- de_res[["qlf_list"]][[donor]]$table
#     tmp$gene <- rownames(de_res[["qlf_list"]][[donor]]$table)
#     ihw_res <- ihw(PValue ~ logCPM, data = tmp, alpha = 0.05)
#     tmp$FDR <- adj_pvalues(ihw_res)
#     tmp <- tmp[tmp$FDR < fdr_thresh,]
#     if (nrow(tmp) > 0.5) {
#         tmp[["donor"]] <- donor
#         df_de_sig_unst <- bind_rows(df_de_sig_unst, tmp)
#     }
# }
# 
# 
# df_donor_n_de <- df_de_sig_unst %>% 
#     group_by(gene) %>% 
#     dplyr::mutate(id = paste0(donor, gene)) %>% distinct(id, .keep_all = TRUE) %>%
#     group_by(donor) %>%
#     summarise(count = n())

```


## Linking pathway results to selection

Not yet implemented

```{r, eval=FALSE, include=FALSE} 
# df_donor_info <- read.table("data/donor_info_070818.txt")
# df_camera_all_unst$classification <- NA
# df_camera_all_unst$classification[
#     df_camera_all_unst$donor %in% df_donor_info$sampleID[df_donor_info$cons_classification == "selection"]
# ] <- "selection"
# df_camera_all_unst$classification[
#     df_camera_all_unst$donor %in% df_donor_info$sampleID[df_donor_info$cons_classification == "undetermined"]
# ] <- "undetermined"
# df_camera_all_unst$classification[
#     df_camera_all_unst$donor %in% df_donor_info$sampleID[df_donor_info$cons_classification == "neutral"]
# ] <- "neutral"
# df_camera_all_unst$classification <- factor(df_camera_all_unst$classification, 
#                                             levels = c("neutral", "undetermined", "selection"))
# 
# df_camera_nsig <- df_camera_all_unst %>% dplyr::filter(collection == "H", FDR < 0.05) %>% 
#     dplyr::filter(geneset %in% c("HALLMARK_E2F_TARGETS", "HALLMARK_G2M_CHECKPOINT", "HALLMARK_MYC_TARGETS_V1")) %>%
#     group_by(geneset, classification, Direction) %>% 
#     dplyr::mutate(id = paste0(donor, geneset, Direction)) %>% distinct(id, .keep_all = TRUE) %>%
#     summarise(n_sig = n())
# df_camera_nsig
# 
# 
# ## test for different proportions of Up/Down sig genesets
# e2f_select <- df_camera_nsig %>% dplyr::filter(geneset == "HALLMARK_E2F_TARGETS") %>% ungroup() %>% dplyr::select(-geneset) %>%
#     spread(classification, n_sig)
# e2f_mat <- as.matrix(e2f_select[, -1])
# rownames(e2f_mat) <- e2f_select[["e2f_select"]]
# fisher.test(e2f_mat)
# g2m_select <- df_camera_nsig %>% dplyr::filter(geneset == "HALLMARK_G2M_CHECKPOINT") %>% ungroup() %>% dplyr::select(-geneset) %>%
#     spread(classification, n_sig)
# g2m_mat <- as.matrix(g2m_select[, -1])
# rownames(g2m_mat) <- g2m_select[["e2f_select"]]
# fisher.test(g2m_mat)
# myc_select <- df_camera_nsig %>% dplyr::filter(geneset == "HALLMARK_MYC_TARGETS_V1") %>% ungroup() %>% dplyr::select(-geneset) %>%
#     spread(classification, n_sig)
# myc_mat <- as.matrix(myc_select[, -1])
# rownames(myc_mat) <- myc_select[["e2f_select"]]
# fisher.test(myc_mat)
# 
# df_camera_all_unst %>% dplyr::filter(collection == "H") %>% 
#     dplyr::filter(geneset %in% c("HALLMARK_E2F_TARGETS", "HALLMARK_G2M_CHECKPOINT", "HALLMARK_MYC_TARGETS_V1")) %>%
#     group_by(geneset, classification, Direction) %>% 
#     dplyr::mutate(id = paste0(donor, geneset, Direction)) %>% distinct(id, .keep_all = TRUE) %>%
#     ungroup() %>%
#     dplyr::mutate(geneset = gsub("_", " ", gsub("HALLMARK_", "", geneset))) %>%
#     ggplot(aes(y = -log10(FDR), x = classification, colour = Direction)) +
#     ggbeeswarm::geom_quasirandom(alpha = 0.7, size = 4) +
#     geom_hline(yintercept = -log10(0.05), linetype = 2) +
#     facet_wrap(~geneset, nrow = 3) +
#     scale_colour_manual(values = c("gray40", "firebrick")) +
#     coord_flip() +
#     theme_bw(14) +
#     xlab("Selection classification") + ylab("-log10(False discovery rate)") +
#     theme(panel.grid.major.x = element_blank(), panel.grid.minor.x = element_blank())
# 
# ggsave("figures/differential_expression/alldonors_camera_enrichment_H_top_genesets_by_selection.png", 
#        height = 7, width = 10)
# ggsave("figures/differential_expression/alldonors_camera_enrichment_H_top_genesets_by_selection.pdf", 
#        height = 7, width = 10)
# ggsave("figures/differential_expression/alldonors_camera_enrichment_H_top_genesets_by_selection.svg", 
#        height = 7, width = 10)
# 
# df_camera_all_unst %>% dplyr::filter(collection == "H", FDR < 0.05) %>% 
#     dplyr::filter(geneset %in% c("HALLMARK_E2F_TARGETS", "HALLMARK_G2M_CHECKPOINT")) %>%
#     group_by(geneset, classification) %>% 
#     dplyr::mutate(id = paste0(donor, geneset)) %>% distinct(id, .keep_all = TRUE) %>%
#     summarise(n_donors = n()) %>% dplyr::arrange(geneset, n_donors) %>% ungroup() %>%
#     dplyr::mutate(geneset = gsub("_", " ", gsub("HALLMARK_", "", geneset))) %>%
#     ggplot(aes(y = n_donors, x = geneset)) +
#     geom_point(alpha = 0.7, size = 4) +
#     ggthemes::scale_colour_tableau() +
#     coord_flip() +
#     theme_bw(14) +
#     xlab("Gene set") + ylab("Number of significant donors")



df_e2f <- df_camera_all_unst %>% 
    dplyr::filter(grepl("HALLMARK_E2F_TARGETS", geneset)) %>%
    dplyr::mutate(id = paste(donor, contrast, sep = ": "), sig = FDR < 0.05)
ggplot(df_e2f, aes(x = Direction, y = -log10(PValue), colour = sig, 
                      label = id)) +
    geom_violin(aes(group = Direction), fill = "gray90", colour = "gray90") +
    geom_point(size = 4) +
    geom_label_repel(show.legend = FALSE, point.padding = 0.5,
                    data = subset(df_e2f, FDR < 0.05)) +
    scale_colour_manual(values = c("gray50", "firebrick"), 
                        label = c("N.S.", "FDR < 5%"), name = "", drop = FALSE) +
    guides(alpha = FALSE) + 
    ggtitle("E2F TARGETS") +
    theme_classic(20)

ggsave("figures/differential_expression/alldonors_camera_H_e2f_targets.png", height = 8.5, width = 10)
ggsave("figures/differential_expression/alldonors_camera_H_e2f_targets.pdf", height = 8.5, width = 10)
ggsave("figures/differential_expression/alldonors_camera_H_e2f_targets.svg", height = 8.5, width = 10)

df_g2m <- df_camera_all_unst %>% 
    dplyr::filter(grepl("HALLMARK_G2M_CHECKPOINT", geneset)) %>%
    dplyr::mutate(id = paste(donor, contrast, sep = ": "), sig = FDR < 0.05)
ggplot(df_g2m, aes(x = Direction, y = -log10(PValue), colour = sig, 
                      label = id)) +
    geom_violin(aes(group = Direction), fill = "gray90", colour = "gray90") +
    geom_point(size = 4) +
    geom_label_repel(show.legend = FALSE, point.padding = 0.5,
                    data = subset(df_g2m, FDR < 0.05)) +
    scale_colour_manual(values = c("gray50", "firebrick"), 
                        label = c("N.S.", "FDR < 5%"), name = "", drop = FALSE) +
    guides(alpha = FALSE) + 
    ggtitle("G2M CHECKPOINT") +
    theme_classic(20)

ggsave("figures/differential_expression/alldonors_camera_H_g2m_checkpoint.png", height = 8.5, width = 10)
ggsave("figures/differential_expression/alldonors_camera_H_g2m_checkpoint.pdf", height = 8.5, width = 10)
ggsave("figures/differential_expression/alldonors_camera_H_g2m_checkpoint.svg", height = 8.5, width = 10)
```


## Compare pathway results to clone prevalence

Hallmark gene sets. Not yet implemented.

```{r, eval=FALSE, include=FALSE}
head(df_camera_nsig)
head(df_camera_all_unst)

head(de_res$qlf_list$euts$coefficients)
head(de_res$qlf_list$zoxy$coefficients)

df_clone_prevalence <- list()
for (don in names(de_res$qlf_list)) {
    df_clone_prevalence[[don]] <- data_frame(
        donor = don,
        clone = rownames(cell_assign_list[[don]]$full_tree$P),
        prevalence = cell_assign_list[[don]]$full_tree$P[, 1])
}
df_clone_prevalence <- do.call("bind_rows", df_clone_prevalence)
    
df_camera_all_unst <- dplyr::mutate(
    df_camera_all_unst, 
    id = paste(donor, collection, geneset, coeff, stat, sep = ".")
)

## plot clone prevalences across donors
df_clone_prevalence %>%
    ggplot(aes(x = prevalence, y = clone)) +
    geom_density_ridges() +
    theme_ridges()


## assign clone prevalences to camera results
df_camera_all_unst <- dplyr::mutate(
    df_camera_all_unst, 
    up_clone = gsub("(clone[0-9])_clone[0-9]", "\\1", coeff),
    dn_clone = gsub("clone[0-9]_(clone[0-9])", "\\1", coeff),
    donor_upclone = paste0(donor, up_clone),
    donor_dnclone = paste0(donor, dn_clone)) 
df_clone_prevalence1 <- dplyr::mutate(df_clone_prevalence,
                                      up_clone = clone,  
                                      clone = NULL,
                                      donor_upclone = paste0(donor, up_clone),
                                      prevalence_upclone = prevalence, 
                                      prevalence = NULL
)
df_clone_prevalence2 <- dplyr::mutate(df_clone_prevalence,
                                      dn_clone = clone,  
                                      clone = NULL,
                                      donor_dnclone = paste0(donor, dn_clone),
                                      prevalence_dnclone = prevalence, 
                                      prevalence = NULL
)
tmp1 <- dplyr::select(df_camera_all_unst, donor, up_clone, dn_clone, id) %>%
    left_join(., df_clone_prevalence1)
tmp2 <- dplyr::select(df_camera_all_unst, donor, up_clone, dn_clone, id) %>%
    left_join(., df_clone_prevalence2)

tmp3 <- dplyr::inner_join(df_camera_all_unst, tmp1, by = "id")
tmp4 <- dplyr::inner_join(tmp3, tmp2, by = "id")
df_camera_all_unst <- tmp4
df_camera_all_unst <- dplyr::mutate(
    df_camera_all_unst, 
    prevalence_relative_up_clone = prevalence_upclone / prevalence_dnclone)

df_camera_all_unst %>% dplyr::filter(collection == "H", FDR < 0.05) %>% 
    dplyr::filter(geneset %in% c("HALLMARK_E2F_TARGETS", "HALLMARK_G2M_CHECKPOINT",
                                 "HALLMARK_MYC_TARGETS_V1",
                                 "HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION",
                                 "HALLMARK_MYC_TARGETS_V2", "HALLMARK_MITOTIC_SPINDLE")) %>%
    dplyr::mutate(geneset = gsub("_", " ", gsub("HALLMARK_", "", geneset))) %>%
    dplyr::mutate(geneset = factor(geneset, levels = 
                                       c("MITOTIC SPINDLE", "MYOGENESIS", "ANGIOGENESIS",
                                         "EPITHELIAL MESENCHYMAL TRANSITION", "MYC TARGETS V2",
                                        "MYC TARGETS V1", "E2F TARGETS", "G2M CHECKPOINT"))) %>%
    ggplot(aes(y = prevalence_upclone, x = geneset, fill = Direction)) +
    geom_boxplot(width = 0.5) +
    # ggthemes::scale_fill_canva(palette = "Surf and turf") +
    scale_fill_manual(values = c("lightgoldenrod1", "sienna1")) +
    coord_flip() +
    theme_ridges(14) +
    guides(fill = guide_legend(reverse = TRUE)) +
    xlab("Gene set") + ylab("Clone prevalence")

ggsave("figures/differential_expression/top_genesets_H_by_direction_prevalence.png", height = 6, width = 10)
ggsave("figures/differential_expression/top_genesets_H_by_direction_prevalence.pdf", height = 6, width = 10)
ggsave("figures/differential_expression/top_genesets_H_by_direction_prevalence.svg", height = 6, width = 10)


df_camera_all_unst %>% dplyr::filter(collection == "H", stat == "logFC", FDR < 0.05) %>% 
    dplyr::filter(geneset %in% c("HALLMARK_E2F_TARGETS", "HALLMARK_G2M_CHECKPOINT",
                                 "HALLMARK_MYC_TARGETS_V1",
                                 "HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION",
                                 "HALLMARK_MYC_TARGETS_V2", "HALLMARK_MITOTIC_SPINDLE")) %>%
    dplyr::mutate(geneset = gsub("_", " ", gsub("HALLMARK_", "", geneset))) %>%
    dplyr::mutate(geneset = factor(geneset, levels = 
                                       c("MITOTIC SPINDLE", "MYOGENESIS", "ANGIOGENESIS",
                                         "EPITHELIAL MESENCHYMAL TRANSITION", "MYC TARGETS V2",
                                        "MYC TARGETS V1", "E2F TARGETS", "G2M CHECKPOINT"))) %>%
    ggplot(aes(y = prevalence_relative_up_clone, x = geneset, fill = Direction)) +
    geom_hline(yintercept = 1) +
    geom_boxplot(width = 0.5) +
    # ggthemes::scale_fill_canva(palette = "Surf and turf") +
    scale_fill_manual(values = c("lightgoldenrod1", "sienna1")) +
    scale_y_log10() +
    coord_flip() +
    theme_ridges(14) +
    guides(fill = guide_legend(reverse = TRUE)) +
    xlab("Gene set") + ylab("Clone prevalence relative to base clone")


df_camera_all_unst %>% dplyr::filter(collection == "H", stat == "logFC", FDR < 0.05) %>% 
    dplyr::filter(geneset %in% c("HALLMARK_E2F_TARGETS", "HALLMARK_G2M_CHECKPOINT",
                                 "HALLMARK_MYC_TARGETS_V1",
                                 "HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION",
                                 "HALLMARK_ANGIOGENESIS", "HALLMARK_MYOGENESIS",
                                 "HALLMARK_MYC_TARGETS_V2", "HALLMARK_MITOTIC_SPINDLE")) %>%
    dplyr::mutate(geneset = gsub("_", " ", gsub("HALLMARK_", "", geneset))) %>%
    dplyr::mutate(geneset = factor(geneset, levels = 
                                       c("MITOTIC SPINDLE", "MYOGENESIS", "ANGIOGENESIS",
                                         "EPITHELIAL MESENCHYMAL TRANSITION", "MYC TARGETS V2",
                                        "MYC TARGETS V1", "E2F TARGETS", "G2M CHECKPOINT"))) %>%
    ggplot(aes(y = prevalence_relative_up_clone, x = geneset, fill = Direction)) +
    geom_hline(yintercept = 1) +
    ggbeeswarm::geom_quasirandom(pch = 21, size = 5) +
    # ggthemes::scale_fill_canva(palette = "Surf and turf") +
    scale_fill_manual(values = c("lightgoldenrod1", "sienna1")) +
    scale_y_log10() +
    coord_flip() +
    theme_ridges(14) +
    guides(fill = guide_legend(reverse = TRUE)) +
    xlab("Gene set") + ylab("Clone prevalence relative to base clone")

df_camera_all_unst %>% dplyr::filter(collection == "H", stat == "logFC", FDR < 0.1) %>% 
    dplyr::filter(geneset %in% c("HALLMARK_E2F_TARGETS", "HALLMARK_G2M_CHECKPOINT",
                                 "HALLMARK_MITOTIC_SPINDLE")) %>%
    group_by(geneset) %>%
  do(broom::tidy(t.test(prevalence_upclone ~ Direction, data = .)))


```



## Write DE and pathway results to file

```{r write-results}
## Write DE results to file:
for (don in names(de_res$qlf_list)) {
  de_res$qlf_list[[don]]$table %>% 
    dplyr::mutate(gene = rownames(.), FDR = IHW::adj_pvalues(IHW::ihw(PValue, logCPM, alpha = 0.1))) %>%
    dplyr::arrange(FDR) %>% write_tsv(
      paste0("output/differential_expression/", don, "_qlf_de_results.tsv"))
}

for (don in names(de_res$camera$H)) {
  for (cntrst in names(de_res$camera$H[[don]])) {
    de_res$camera$H[[don]][[cntrst]]$logFC %>% 
      dplyr::mutate(geneset = rownames(.)) %>%
      dplyr::arrange(FDR) %>% write_tsv(
        paste0("output/differential_expression/", don, "_camera_hallmark_geneset_results_", cntrst, ".tsv"))    
  }
}

```