---
title: "DEWSeq Analysis"
author: "Thomas Schwarzl and Sudeep Sahadevan"
date: "07/10/2020"
output:
  BiocStyle::html_document:
    toc: true
    toc_float: true
    toc_depth: 5
params:
  protein: ""
  sampleinfo_file: ""
  countmatrix_file: ""
  annotation_file: ""
  output_windows_file: ""
  output_regions_file: ""
  output_bed_file: ""
  output_rdata: ""
  min_count: 2
  min_sample: 2
  LRT: FALSE
  p_value_cutoff: 0.05
  lfc_cutoff: 1
  overlap_correction: FALSE
  IHW: TRUE
  decide_fit: TRUE
---
```{r setup,  include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```


```{r assignData,  echo=FALSE,  eval=TRUE}
protein <- params$protein
sampleinfo_file <- params$sampleinfo_file
countmatrix_file <- params$countmatrix_file
annotation_file <- params$annotation_file
output_windows_file <- params$output_windows_file
output_regions_file <- params$output_regions_file
output_bed_file <- params$output_bed_file
output_rdata <- params$output_rdata
min_count <- params$min_count
min_sample <- params$min_sample
p_value_cutoff <- params$p_value_cutoff
lfc_cutoff <- params$lfc_cutoff
```

Sanity check input parameter values

```{r sanityCheck,  echo=FALSE,  eval=TRUE}
# round of min_count and min_sample values first
message("Any decimals given as values for min_count and min_sample parameters will be rounded off to the nearest integer.")
min_count <- round(min_count)
min_sample <- round(min_sample)
# receive and sanity check p_value_cutoff
if (p_value_cutoff <= 0 || p_value_cutoff >= 1) {
  warning("p_value_cutoff must satisfy: 0<=p_value_cutoff<=1. Resetting to default value: 0.05")
  p_value_cutoff <- 0.05
}
# sanity check log2Foldchange cutoff
if (lfc_cutoff < 0) {
  warning("lfc_cutoff must be a value >=0. Resetting to default value: 1.00")
  lfc_cutoff <- 1.0
}
# sanity check LRT vs Wald
if (is(params$LRT,  "logical")) {
  lrt <- params$LRT
}else {
  warning("LRT must be TRUE or FALSE,   setting this parameter to default: FALSE")
  lrt <- FALSE
}
# sanity check overlap correction parameter
if (is(params$overlap_correction,  "logical")) {
  overlap_correction <- params$overlap_correction
}else {
  warning("overlap_correction must be TRUE or FALSE,  setting this parameter to default: TRUE")
  overlap_correction <- TRUE
}
# sanity check IHW vs BH correction parameter
if (is(params$IHW, "logical")) {
  ihw_filt <- params$IHW
}else {
  warning("IHW must be TRUE or FALSE,  setting this parameter to default: TRUE")
  ihw_filt <- TRUE
}
# sanity check automated fit vs parametric fit paramter
if (is(params$decide_fit, "logical")) {
  decide_fit <- params$decide_fit
}else {
  warning("decide_fit must be TRUE or FALSE,  setting this parameter to default: TRUE")
  decide_fit <- TRUE
}
```

# eCLIP analysis of `r protein` 


## Setup

This is the analysis of __`r protein`__ with    
sampleinfo file: ``r sampleinfo_file``   
countmatrix file: ``r countmatrix_file`` and     
annotation file: ``r annotation_file`` 

with the following threshold:

minimum read count per window per sample: ``r min_count``
number of samples with minimum read count per window: ``r min_sample``

using the following parameters:   
p-value cut-off: ``r p_value_cutoff``  
Log2FoldChange cut-off: ``r lfc_cutoff``
use automated method for dispersion estmation: ``r decide_fit``    
use LRT test : ``r lrt``  
use overlap correction: ``r overlap_correction``   
use IHW for FDR correction: ``r ihw_filt``   


``` {r check if files exists,  echo=FALSE}
file_exists <- function(x) {
  if (!file.exists(x))
    stop(paste0("'",  x,  "' - file does not exist."))
}

file_exists(countmatrix_file)
file_exists(annotation_file)
file_exists(sampleinfo_file)
```

First,  we load the libraries.

```{r load DEWSeq}
required_packages <- c("DEWSeq", "data.table", "IHW", "R.utils", "tidyverse", "ggrepel")
installed_packages <- installed.packages()[, 1]
missing_packages <- setdiff(required_packages, installed_packages)
if (length(missing_packages) != 0) {
  stop("Found missing dependencies! Please install the following package(s): ", paste(missing_packages, collapse = ", "))
}
suppressPackageStartupMessages({
  require(DEWSeq)
  require(tidyverse)
  require(data.table)
  require(IHW)
  require(R.utils)
  require(ggrepel)
})
```

## Read in data

Here we read in the window counts

```{r read window counts}
window_counts <- fread(countmatrix_file,  sep = "\t", stringsAsFactors = FALSE) %>% as.data.frame()
rownames(window_counts) <- window_counts[, 1]
window_counts <- window_counts[, -1]
```

and the sample info file

```{r read sample info}
sample_info <- read.table(sampleinfo_file, sep = "\t", stringsAsFactors = FALSE)
if (ncol(sample_info) < 2) {
  stop("sampleinfo_file ", sampleinfo_file, " MUST have atleast two columns: first column should be the sample names used in ", countmatrix_file,
       " and second column must be the experiment type: IP or SMI")
}else if (ncol(sample_info) > 2) {
  message("Found ", ncol(sample_info), " columns in ", sampleinfo_file, " using the first column as sample name and second column as experiment name")
  sample_info <- sample_info[, c(1, 2)]
}
colnames(sample_info) <- c("samples", "type")
rownames(sample_info) <- sample_info[, 1]
```

Now we make sure that the sampleinfo file contains the column "type" with values "SMI" and "IP" only. 

```{r sampleSanity}
# make sure that sample_info rows and window_counts columns are in same order
common_samples <- sort(intersect(colnames(window_counts), rownames(sample_info)))
if (length(common_samples) != ncol(window_counts)) {
  stop("The number of samples in ", countmatrix_file, " and ", sampleinfo_file, " do not MATCH!")
}
sample_info <- sample_info[common_samples,  ]
window_counts <- window_counts[, common_samples]
# Now make sure that sample_info$type contains only "IP" and "SMI"
type_check <- setdiff(unique(sample_info$type), c("IP", "SMI"))
if (length(type_check) != 0) {
  stop("The second column in ", sampleinfo_file, " should contain analysis types:'IP' or 'SMI' only. Found unknown value(s): ",
       paste(type_check, collapse = ",  "))
}
```

We make sure that only IP and SMI are in the right factor level order

```{r sampleFactor}
sample_info <- sample_info %>% mutate(type = factor(type,  levels = c("SMI",  "IP")))
```



We create the DEWSeq object

```{r dewseqInit}
ddw <- DESeqDataSetFromSlidingWindows(countData  = window_counts,
                                      colData    = sample_info,
                                      annotObj   = annotation_file,
                                      tidy       = FALSE,
                                      design     = ~type)
```


## Prefiltering

```{r prefiltering1}
# remove all empty windows
keep <- rowSums(counts(ddw)) >= 1
ddw <- ddw[keep, ]
```


## Estimating size factors


```{r size factors}
ddw <- estimateSizeFactors(ddw)
sizeFactors(ddw)
```

### estimate size factors for only protein_coding genes

```{r protein_coding_size_factors}
ddw_mrnas <- ddw[rowData(ddw)[, "gene_type"] == "protein_coding",  ]
ddw_mrnas <- estimateSizeFactors(ddw_mrnas)
```

### estimate size factors without significant windows

```{r size_factors_no_sig_windows}
ddw_tmp <- ddw
ddw_tmp <- estimateDispersions(ddw_tmp,  fitType = "local",  quiet = TRUE)
if (lrt) {
  ddw_tmp <- nbinomLRT(ddw_tmp, full = ~type, reduced = ~1)
}else {
  ddw_tmp <- nbinomWaldTest(ddw_tmp)
}

tmp_significant_windows <-
                results(ddw_tmp,
                    contrast = c("type",  "IP",  "SMI"),
                    tidy = TRUE,
                    filterFun = ihw) %>%
                dplyr::filter(padj < p_value_cutoff) %>%
                .[["row"]]
rm(ddw_tmp)
```

estimate the size factors without the significant windows.

```{r final_size_factors}
ddw_mrnas <- ddw_mrnas[!rownames(ddw_mrnas) %in% tmp_significant_windows,  ]
ddw_mrnas <- estimateSizeFactors(ddw_mrnas)
```

before thresholding:
```{r threshold1}
dim(ddw)
```

Now threshold the windows read count table.
```{r threshold2}
keep_exp <-  which(rowSums(counts(ddw) > min_count) >= min_sample)
ddw <- ddw[keep_exp, ]
```

after thresholding:
```{r threshold3}
dim(ddw)
```
assign size factors

```{r final_assign}
sizeFactors(ddw) <- sizeFactors(ddw_mrnas)
rm(list = c("tmp_significant_windows",  "ddw_mrnas"))
sizeFactors(ddw)
```


## Differential window analysis

### Dispersion estimates

```{r source,  echo = FALSE,  eval = FALSE}
# source: https://support.bioconductor.org/p/81094/
```
We fit parametric and local fit,  and decide the best fit following this [Bioconductor post](https://support.bioconductor.org/p/81094/)

```{r parametric_dispersion}
parametric_ddw  <- estimateDispersions(ddw,  fitType = "parametric")
if (decide_fit) {
  local_ddw  <- estimateDispersions(ddw,  fitType = "local")
}

```

This is the dispersion estimate for parametric fit
```{r plot parametric fit,  fig.wide=TRUE}
plotDispEsts(parametric_ddw,  main = "Parametric fit")
```

This is the dispersion estimate for local fit,  given automated decision fitting is enabled:
```{r plot local fit,  fig.wide = TRUE}
if (decide_fit) {
  plotDispEsts(local_ddw,  main = "Local fit")
}

```

This will get the residuals for either fit,  only for automated decision fitting
```{r residual fit}
parametric_resid <- na.omit(with(mcols(parametric_ddw), abs(log(dispGeneEst) - log(dispFit))))
if (decide_fit) {
  local_resid <- na.omit(with(mcols(local_ddw), abs(log(dispGeneEst) - log(dispFit))))
  resid_df <- data.frame(residuals = c(parametric_resid, local_resid),
                         fitType = c(rep("parametric", length(parametric_resid)),
                                   rep("local", length(local_resid))))
  summary(resid_df)
}

```

and we plot histograms of the fits

```{r plot residual histograms,  fig.wide = TRUE}
if (decide_fit) {
  ggplot(resid_df, aes(x = residuals, fill = fitType)) +
    scale_fill_manual(values = c("darkred", "darkblue")) + geom_histogram(alpha = 0.5, position = "identity", bins = 100) +
    theme_bw()
}

```

Now,  we will decide for the better fit based on median

```{r choose_fit}
summary(parametric_resid)
if (decide_fit) {
  summary(local_resid)
  if (median(local_resid) <= median(parametric_resid)) {
    cat("chosen fitType: local")
    ddw <- local_ddw
  }else {
    cat("chosen fitType: parametric")
    ddw <- parametric_ddw
  }
  rm(local_ddw, parametric_ddw, resid_df, parametric_resid, local_resid)
}else {
  ddw <- parametric_ddw
  rm(parametric_ddw)
}

```



### Wald test or LRT

```{r wald_or_LRT}
if (lrt) {
  ddw <- nbinomLRT(ddw, full = ~type,  reduced = ~1)
}else {
  ddw <- nbinomWaldTest(ddw)
}

```


### Significance testing

```{r sig_windows}
result_windows <- resultsDEWSeq(ddw,
                              contrast = c("type",  "IP",  "SMI"),
                              tidy = TRUE) %>% as_tibble

result_windows
```


### Multiple hypothesis correction with IHW

You might be interested to correct for multiple hypothesis testing with IHW.

Decide on overlap correction based on the parameter `overlap_correction`

```{r ihw}
if (overlap_correction && ihw_filt) {
  result_windows[, "p_adj_IHW"] <- adj_pvalues(ihw(pSlidingWindows ~ baseMean,
                     data = result_windows,
                     alpha = p_value_cutoff,
                     nfolds = 10))
  padj_col <- "p_adj_IHW"
}else if (!overlap_correction && ihw_filt) {
  result_windows[, "p_adj_IHW"] <- adj_pvalues(ihw(pvalue ~ baseMean,
                     data = result_windows,
                     alpha = p_value_cutoff,
                     nfolds = 10))
  padj_col <- "p_adj_IHW"
}else if (overlap_correction && !ihw_filt) {
  padj_col <- "pSlidingWindows.adj"
}else {
  result_windows[, "p_adj"] <- p.adjust(result_windows$pvalue, method = "BH")
  padj_col <- "p_adj"
}

```

Determine significant windows

```{r filter_sig_windows}
result_windows <- result_windows %>%
                      mutate(significant = result_windows[, padj_col] < p_value_cutoff)
sig_windows <- sum(result_windows$significant)
```


`r sig_windows` windows are significant


### Combining windows

```{r,  reg1,  message=FALSE,  eval=TRUE, include=FALSE}
if (sig_windows > 0) {
 result_regions <- extractRegions(windowRes  = result_windows,  padjCol    = padj_col,  padjThresh = p_value_cutoff,  log2FoldChangeThresh = lfc_cutoff) %>% as_tibble
}else {
 message("Cannot find significant windows in this dataset. Try lowering the p-value and log2FoldChange thresholds!")
}
```

```{r extractRegion, eval=FALSE}
if (sig_windows > 0) {
 result_regions <- extractRegions(windowRes  = result_windows,  padjCol    = padj_col,  padjThresh = p_value_cutoff,  log2FoldChangeThresh = lfc_cutoff) %>% as_tibble
}
```

                     
### Writing Bed file

```{r writing bed file}
if (sig_windows > 1) {
 toBED(windowRes = result_windows,  regionRes = result_regions, padjThresh = p_value_cutoff, padjCol   = padj_col,  fileName  = output_bed_file)
}else {
 message("This analysis does not have enough <=1 significant windows to create BED file for visualization")
}
```


## Save Session
```{r save_data}
# save enriched windows,  gzip results file if the file suffix is .gz
if (grepl("\\.gz$", output_windows_file, ignore.case = TRUE)) {
  gz_out <- gzfile(output_windows_file, "w")
  write.table(result_windows, file = gz_out, sep = "\t", quote = FALSE, row.names = FALSE, col.names = TRUE)
  close(gz_out)
}else {
  write.table(result_windows, file = output_windows_file, sep = "\t", quote = FALSE, row.names = FALSE, col.names = TRUE)
}
# save enriched regions
if (sig_windows > 0) {
  if (grepl("\\.gz$", output_regions_file, ignore.case = TRUE)) {
    gz_out <- gzfile(output_regions_file, "w")
    write.table(result_regions, file = gz_out, sep = "\t", quote = FALSE, row.names = FALSE, col.names = TRUE)
    close(gz_out)
  }else {
    write.table(result_regions, file = output_regions_file, sep = "\t", quote = FALSE, row.names = FALSE, col.names = TRUE)
  }
}
# save session
# Warning! session images can be heavy!
if (nchar(output_rdata) > 5) {
  save.image(file = output_rdata)
}
```

## Session Info

```{r session info}
sessionInfo()
```