step7-regMatVsDERs.Rmd

---
output:
  html_document:
    toc: true
    toc_float: true
    code_folding: hide
    theme: spacelab
---

Compare output from regionMatrix() with DERs
============================================

# Setup

## Libraries

```{r 'libraries', dev = 'CairoPNG'}
library('GenomicRanges')
library('DT')
library('knitr')
library('ggplot2')
library('devtools')
opts_chunk$set(dev = 'CairoPNG')
```


## Load data

```{r 'data', bootstrap.show.output = FALSE}
## Check that data was loaded
stopifnot(all(c('fullRegions', 'regionMat', 'analysisPath') %in% ls()))

## Fix DERs
ders <- fullRegions
names(ders) <- NULL

## Fix region matrices
regs <- unlist(GRangesList(lapply(regionMat, '[[', 'regions')))
names(regs) <- NULL

## Assign seqlengths
data(hg19Ideogram, package = 'biovizBase', envir = environment())
seqlengths(ders) <- seqlengths(hg19Ideogram)[names(seqlengths(ders))]
seqlengths(regs) <- seqlengths(hg19Ideogram)[names(seqlengths(regs))]

## Sort
ders <- sort(ders)
regs <- sort(regs)

## Check names match
identical(seqlengths(regs), seqlengths(ders))

## Save the raw data
save(ders, file = file.path(analysisPath, 'dersOriginal.Rdata'))
save(regs, file = file.path(analysisPath, 'regsOriginal.Rdata'))

## Filter out those with less than 3bp
c('# ders under 3bp' = sum(width(ders) < 3), '# regs under 3bp' = sum(width(regs) < 3))
ders <- ders[width(ders) >= 3]
regs <- regs[width(regs) >= 3]

## Save data used
save(ders, file = file.path(analysisPath, 'ders.Rdata'))
save(regs, file = file.path(analysisPath, 'regs.Rdata'))
```

Construct logical indexes for DERs and regionMatrix regions.

```{r 'buildIndex'}
## Construct logical Rle indexes for bases with some region
build_index <- function(gr) {
    res <- lapply(names(seqlengths(gr)), function(chr) {
        chr.len <- seqlengths(gr)[chr]
        ir <- sort(ranges(gr[seqnames(gr) == chr]))
        log <- c(rep(c(FALSE, TRUE), length(ir)), FALSE)
    
        starts <- ends <- rep(NA, length(ir) * 2)
        i <- rep(c(TRUE, FALSE), length(ir))
        starts[i] <- start(ir)
        ends[i] <- end(ir)
    
        starts[!i] <- ends[i] + 1
    
        if(max(ends, na.rm = TRUE) < chr.len) {
            ends[!i] <- c(starts[i] - 1, chr.len)[-1]
        } else {
            ends[!i] <- c(starts[i] - 1, NULL)[-1]
            starts <- starts[- length(starts)]
            log <- log[- length(log)]
        }
   
        if(starts[1] != 1) {
            ends <- c(starts[1] - 1, ends)
            starts <- c(1, starts)
        } else {
            log <- log[-1]
        }
   
        widths <- mapply(function(s, e) { e - s + 1}, starts, ends)
    
        Rle(log, widths)
    })
    names(res) <- names(seqlengths(gr))
    return(res)
}
index.ders <- build_index(ders)
index.regs <- build_index(regs)

## Add info for chrs where there are no regs
miss <- !paste0('chr', c(1:22, 'X', 'Y')) %in% names(index.regs)
names(miss) <- paste0('chr', c(1:22, 'X', 'Y'))
if(any(miss)) {
    miss.add <- lapply(names(miss)[miss], function(x) { 
        Rle(FALSE, seqlengths(hg19Ideogram)[x]) 
    })
    names(miss.add) <- names(miss)[miss]
    index.regs <- c(index.regs, miss.add)
    index.regs <- index.regs[match(names(miss), names(index.regs))]
}

## Add info for chrs where there are no DERs
miss <- !names(index.regs) %in% names(index.ders)
if(any(miss)) {
    miss.add <- lapply(names(index.regs)[miss], function(x) { 
        Rle(FALSE, length(index.regs[[x]])) 
    })
    names(miss.add) <- names(index.regs)[miss]
    index.ders <- c(index.ders, miss.add)
    index.ders <- index.ders[match(names(index.regs), names(index.ders))]
}
```

# Compare

## Visually explore

```{r 'epivizr', eval = FALSE}
library('epivizr')
mgr <- startEpiviz()
ders_dev <- mgr$addDevice(ders[!as.logical(ders$significantFWER)], "DERs no sig FWER")
ders_sig_dev <- mgr$addDevice(ders[as.logical(ders$significantFWER)], "DERs sig FWER")
regs_dev <- mgr$addDevice(regs, "Region Matrix")

## SOX11
mgr$navigate("chr2", 5810000, 5850000)

## MEX3A
mgr$navigate("chr1", 156040000, 156090000)

## VASH2
mgr$navigate("chr1", 213120000, 213170000)

## TG:
mgr$navigate("chr8", 134040000, 134120000)

## IGF2BP2
mgr$navigate("chr3", 185350000, 185410000)

## FBN3
mgr$navigate("chr19", 8130000, 8180000)

## End
mgr$stopServer()
```


## Basic comparison

Number of regions

```{r 'basic1'}
## Number of regions
c('ders' = length(ders), 'regs' = length(regs))
```

Summary on width of regions

```{r 'basic2'}
## Size of regions
c('ders' = summary(width(ders)), 'regs' = summary(width(regs)))
```


## Compare indexes

### Base-pairs

Number of base-pairs in each index. Summary first, then overall info for the genome (in number of bases, then in percent of the genome), and finally results in interactive table.

```{r 'index-num'}
## Merge all the indexes
index.all <- mapply(function(der, reg) {
    both <- der & reg
    only.der <- der & !reg
    only.reg <- !der & reg
    none <- !der & !reg
    
    res <- list('both' = both, 'only.der' = only.der, 'only.reg' = only.reg,
        'none' = none, 'all.der' = der, 'all.reg' = reg)
    return(list(res))
}, index.ders, index.regs)

## Find number of base-pairs in each index
index.num <- data.frame(do.call(rbind, lapply(index.all, function(x) { sapply(x, sum)})))
index.num$chrLen <- seqlengths(ders)
index.num$chr <- rownames(index.num)
rownames(index.num) <- NULL

## Print info
summary(index.num)

## Overall info
overallInfo <- colSums(index.num[, -ncol(index.num)])
overallInfo

## Overall info in percent
overallInfo / sum(as.numeric(index.num$chrLen)) * 100
```

```{r 'print-index-num', results = 'asis'}
d1 <- data.frame(row = seq_len(nrow(index.num)), index.num, check.names=FALSE)
datatable(d1, options = list(pagingType='full_numbers', pageLength=25, scrollX='100%'), rownames = FALSE)
```

### Segments per index

Number of segments per index. First summary, then results for genome, and finally an interactive table.


```{r 'index-seg'}
## Find number of segments in each index
index.seg <- data.frame(do.call(rbind, lapply(index.all, function(x) { 
    sapply(x, function(y) {
        sum(runValue(y))
    })
})))
index.seg$chr <- rownames(index.seg)
rownames(index.seg) <- NULL

## Print info
summary(index.seg)

## Overall info
colSums(index.seg[, -ncol(index.seg)])
```


```{r 'print-index-seg', results = 'asis'}
d2 <- data.frame(row = seq_len(nrow(index.seg)), index.seg, check.names=FALSE)
datatable(d2, options = list(pagingType='full_numbers', pageLength=25, scrollX='100%'), rownames = FALSE)
```


### Segments width

Summary of the segment widths for each index. First the overall summary, then the results for each index.

```{r 'index-width'}
## Get an idea of the width of the segments in each index
index.width <- data.frame(do.call(rbind, lapply(index.all, function(x) {
    tmp <- data.frame(do.call(rbind, lapply(x, function(y) {
        summary(runLength(y)[runValue(y)])
    })), check.names = FALSE)
    tmp$index <- names(x)
    rownames(tmp) <- NULL
    return(tmp)
})), check.names = FALSE)
index.width$chr <- rep(names(seqlengths(ders)), each = 6)
rownames(index.width) <- NULL

## Print info
summary(index.width)
```

```{r 'print-index-width', results = 'asis'}
d3 <- data.frame(row = seq_len(nrow(index.width)), index.width, check.names=FALSE)
datatable(d3, options = list(pagingType='full_numbers', pageLength=25, scrollX='100%'), rownames = FALSE)
```


## Overlaps

### Minimum 20 bp

```{r 'overlaps20'}
ov20 <- findOverlaps(ders, regs, minoverlap = 20L)

counts <- list()
for(type in c("any", "within", "equal")) {
    ct.ders <- countOverlaps(ders, regs, minoverlap = 20L, type = type)
    plot(log10(table(ct.ders)), main = paste("DERs in regs for type", type))
    ct.regs <- countOverlaps(regs, ders, minoverlap = 20L, type = type)
    plot(log10(table(ct.regs)), main = paste("Regs in DERs for type", type))
    counts <- c(counts, list(table(ct.ders), table(ct.regs)))
}
cts <- as.integer(unique(unlist(lapply(counts, names))))

nOverlap20 <- do.call(rbind, lapply(counts, function(x) { 
    df <- data.frame(nOverlap = cts, freq = x[match(cts, names(x))],
        row.names = seq_len(length(cts)))
    df$observed <- !is.na(df$freq)
    df$freq[is.na(df$freq)] <- 0
    df$cumFreq <- cumsum(df$freq)
    df$cumPerc <- df$cumFreq / max(df$cumFreq) * 100
    return(df)
}))
nOverlap20$type <- factor(rep(c('any', 'within', 'equal'), each = length(cts) * 2), levels = c('any', 'within', 'equal'))
nOverlap20$match <- rep(rep(c('DERs-in-regs', 'regs-in-DERs'), each = length(cts)), 3)
#nOverlap <- nOverlap[complete.cases(nOverlap), ]
rownames(nOverlap20) <- NULL

nOverlap20$alpha <- ifelse(nOverlap20$observed, 1, 1/3)
```

Summary plots showing cumulative frequency and cumulative percent.

```{r 'gplot20'}
## Make a nice plot
ggplot(data = nOverlap20, aes(x = nOverlap, y = cumFreq, colour = match, alpha = alpha)) + geom_point() + facet_grid( . ~ type )# + geom_smooth(se=FALSE)

## Show cumulative percents
ggplot(data = nOverlap20, aes(x = nOverlap, y = cumPerc, colour = match, linetype = match)) + geom_line(lwd=1) + facet_grid( . ~ type )
```


Some important numbers: percent of regions with width < 20 bp, base level agreement, region level agreement (min overlap 20 bp).

```{r}
## Percent with widths < 20L
small <- c('ders' = sum(width(ders) < 20) / length(ders), 'regs' = sum(width(regs) < 20) / length(regs)) * 100
data.frame('under-20' = small, '20-and-above' = 100 - small, check.names = FALSE)

## Base-level agreement
c('regs' = overallInfo['both'] / (overallInfo['both'] + overallInfo['only.reg']) * 100, 'ders' = overallInfo['both'] / (overallInfo['both'] + overallInfo['only.der']) * 100)

## Overlap (min 20) agreement
c('regs' = 100 - subset(nOverlap20, match == 'regs-in-DERs' & nOverlap == 0 & type == 'any')$cumPerc, 'ders' = 100 - subset(nOverlap20, match == 'DERs-in-regs' & nOverlap == 0 & type == 'any')$cumPerc)
```

### Minimum 1 bp

```{r 'overlaps1'}
ov1 <- findOverlaps(ders, regs, minoverlap = 1L)

counts <- list()
for(type in c("any", "within", "equal")) {
    ct.ders <- countOverlaps(ders, regs, minoverlap = 1L, type = type)
    plot(log10(table(ct.ders)), main = paste("DERs in regs for type", type))
    ct.regs <- countOverlaps(regs, ders, minoverlap = 1L, type = type)
    plot(log10(table(ct.regs)), main = paste("Regs in DERs for type", type))
    counts <- c(counts, list(table(ct.ders), table(ct.regs)))
}
cts <- as.integer(unique(unlist(lapply(counts, names))))

nOverlap1 <- do.call(rbind, lapply(counts, function(x) { 
    df <- data.frame(nOverlap = cts, freq = x[match(cts, names(x))], 
        row.names = seq_len(length(cts)))
    df$observed <- !is.na(df$freq)
    df$freq[is.na(df$freq)] <- 0
    df$cumFreq <- cumsum(df$freq)
    df$cumPerc <- df$cumFreq / max(df$cumFreq) * 100
    return(df)
}))
nOverlap1$type <- factor(rep(c('any', 'within', 'equal'), each = length(cts) * 2), levels = c('any', 'within', 'equal'))
nOverlap1$match <- rep(rep(c('DERs-in-regs', 'regs-in-DERs'), each = length(cts)), 3)
#nOverlap <- nOverlap[complete.cases(nOverlap), ]
rownames(nOverlap1) <- NULL

nOverlap1$alpha <- ifelse(nOverlap1$observed, 1, 1/3)

## Overlap (min 1bp) agreement
c('regs' = 100 - subset(nOverlap1, match == 'regs-in-DERs' & nOverlap == 0 & type == 'any')$cumPerc, 'ders' = 100 - subset(nOverlap1, match == 'DERs-in-regs' & nOverlap == 0 & type == 'any')$cumPerc)
```

Summary plots showing cumulative frequency and cumulative percent.

```{r 'gplot1'}
## Make a nice plot
ggplot(data = nOverlap1, aes(x = nOverlap, y = cumFreq, colour = match, alpha = alpha)) + geom_point() + facet_grid( . ~ type )# + geom_smooth(se=FALSE)

## Show cumulative percents
ggplot(data = nOverlap1, aes(x = nOverlap, y = cumPerc, colour = match, linetype = match)) + geom_line(lwd=1) + facet_grid( . ~ type )
```

# Save results

```{r 'save'}
save(index.all, index.num, index.seg, index.width, nOverlap20, ov20, nOverlap1, ov1, overallInfo, file = file.path(analysisPath, "comparison-results.Rdata"))
```





# Reproducibility

Analysis path: `r  analysisPath`

Re-make the report

```{r 'remake', eval = FALSE}
# Load fullRegions.Rdata and regionMat.Rdata before this step
library('rmarkdown')
render('step7-regMatVsDERs.Rmd')
```


Date the report was generated.

```{r reproducibility1, echo=FALSE}
## Date the report was generated
Sys.time()
```

`R` session information.

```{r reproducibility3, echo=FALSE}
## Session info
options(width = 120)
session_info()
```