「R绘图」minimap2的PAF文件如何进行可视化？

Minimap2是知名比对工具BWA的开发者Li Heng新开发的比对工具，它能够快速的将DNA或者mRNA序列比对到参考基因组上，使用场景有下面几种：

• 将PacBio或OXford Nanopore的read和已有参考基因组（如人类）进行比对
• 寻找高错误率read(15%)之间的overlap
• 将PacBio Iso-Seq 或Nanopore cDNA或RNA序列比对到参考基因组
• 将illumina 单端或者双端序列比对到参考基因组
• 组装之间的比对
• 临近物种的全基因组比对

最近用Canu的不同参数装了几个版本的基因组，希望比较不同组装之间是否连续，要是在组装结果A里面的是ABCD的排布，在结果B里面却出现了倒置变成了ACBD，那么我就要怀疑人生了。

我仔细看了一下PAF的输出格式，发现输出结果非常友好，但是不可能直接用read.table或者data.table::fread读取，所以我就自己写了一个解析函数。同时为了提高我对R语言中grid系统的理解，我又写了一个专门的画图代码

代码见最后，100多行的样子，保存为paf.R, 使用方法为:

source(".paf.R")
df <- read.paf("./asm_vs_ref.paf")
plot_synteny(df)

两个临近物种的结果图如下

效果图1局部

同一物种不同组装的效果图

效果图2局部

我原本以为画这种图需要累加坐标，后来发现原来grid系统里面的视图(viewport)可以继续分为很多个图层，每个图层可有不同的坐标系统。

# paf.R
read.paf <- function(file, sep = "\t",
                     header = FALSE,
                     MQ = 10,
                     ...){
  data <- readLines(file)
  
  dataSize <- length(data)
  # initialize 
  qName   <- vector("character", dataSize)
  qLength <- vector("integer", dataSize)
  qStart  <- vector("integer", dataSize)
  qEnd    <- vector("integer", dataSize)
  strand  <- vector("character", dataSize)
  tName   <- vector("character", dataSize)
  tLength <- vector("integer", dataSize)
  tStart  <- vector("integer", dataSize)
  tEnd    <- vector("integer", dataSize)
  reMatch <- vector("integer", dataSize)
  bLength <- vector("integer", dataSize)
  mQuality<- vector("integer", dataSize)

  i <- 1
  j <- 1
  for (i in seq(dataSize)){
    items   <- strsplit(data[i], split = sep)[[1]]
    
    quality <- as.numeric(items[12]) #Mapping Quality 0-255
    if (quality < MQ)
      next
    qName[j]   <- items[1] # Query sequence name
    qLength[j] <- as.integer(items[2])
    qStart[j]  <- as.integer(items[3]) + 1L
    qEnd[j]    <- as.integer(items[4]) + 1L # convert 0-based to 1-based
    strand[j]  <- items[5] # Relative strand: "+" or "-"
    tName[j]   <- items[6] # Target sequence name
    tLength[j] <- as.integer(items[7])
    tStart[j]  <- as.integer(items[8]) + 1L
    tEnd[j]    <- as.integer(items[9]) + 1L
    reMatch[j] <- as.integer(items[10])  #Number of residue matches
    bLength[j] <- as.integer(items[11]) #Alignment block length
    mQuality[j]<- as.integer(items[12]) #Alignment block length
    j <- j + 1
    
    pafDataframe <- data.frame(qName = qName[1:j], qStart = qStart[1:j], qEnd = qEnd[1:j],
                               tName = tName[1:j], tStart = tStart[1:j], tEnd = tEnd[1:j],
                               qLength = qLength[1:j], tLength = tLength[1:j], 
                               strand  = strand[1:j],
                               reMatch = reMatch[1:j], bLength = bLength[1:j],
                               mQuality = mQuality[1:j],
                               stringsAsFactors = FALSE)
  }
  return(pafDataframe) 
}


plot_synteny <- function(df, contigs = 20,
                         lineSize = 3,
                         borderCol = "#5496ff",... ){
  
  # select the top N contig
  x <- df[,c("qName","qLength")]
  x <- x[!duplicated(x$qName),]
  x <- x[order(x$qLength, decreasing = TRUE),][1:contigs,]

  y <- df[,c("tName","tLength")]
  y <- y[!duplicated(y$tName),]
  y <- y[order(y$tLength, decreasing = TRUE),][1:contigs,]
  
  # make new page for ploting
  grid::grid.newpage()
  
  # allot the ratio of each contig
  x_frac <- x$qLength / sum(x$qLength)
  y_frac <- y$tLength / sum(y$tLength)
  
  # draw the contig name
  grid::pushViewport(grid::viewport(height = 0.7,
                        width =  0.7,
                        gp = grid::gpar(cex = 0.75),
                        name = "contigName"))
  x_pos <- c(0, cumsum(x_frac)[1:(contigs-1)])
  y_pos <- 1- c(0, cumsum(y_frac)[1:(contigs-1)])
  for (i in seq.int(1,contigs)){
    grid::grid.text(label = x[i,1], 
              x = grid::unit(x_pos[i], "npc"),
              y = grid::unit(1, "npc"),
              just = c("left","bottom"),
              rot = 35
              )
    grid::grid.text(label = y[i, 1],
              x = grid::unit(1, "npc"),
              y = grid::unit(y_pos[i], "npc"),
              just = c("left")
              )
    
  }
  
  vplay <- grid::grid.layout(contigs, contigs, 
                       widths  = x_frac,
                       heights = y_frac)
  grid::pushViewport(grid::viewport(layout = vplay,
                        name = "vplay"))
  
  
  # line represent the synteny
  for ( i in seq(1, contigs)){
    for (j in seq(1, contigs)){
      
      # get the name and length for subsetting
      xName   <- x[i,1]
      xLength <- x[i,2]
      yName   <- y[j,1] 
      yLength <- y[j,2]
      
      # push view port for plot Collinearity
      grid::pushViewport(grid::viewport(layout.pos.col = i,
                            layout.pos.row = j,
                            xscale = c(1, xLength),
                            yscale = c(1, yLength),
                            name = paste0("pos",i,j)))
  
      grid::grid.rect(gp=grid::gpar(col=borderCol))
      # select the data
      plot_df <- df[df$qName == xName & df$tName == yName,] 
      blocks <- nrow(plot_df)
      #cat(sprintf("block size is %d\n", blocks))
      
      if (blocks == 0) {
        grid::upViewport() 
        next
      }
      
      # plot the line
      for (k in seq(1, blocks)){
        if (plot_df$strand[k] == "+"){
          grid::grid.lines(x = c(plot_df$qStart[k], plot_df$qEnd[k]),
                     y = c(plot_df$tStart[k], plot_df$tEnd[k]),
                     gp=grid::gpar(lwd = lineSize),
                     default.units = "native")
        } else {
          grid::grid.lines(x = c(plot_df$qStart[k], plot_df$qEnd[k]),
                     y = c(plot_df$tEnd[k], plot_df$tStart[k]),
                     gp=grid::gpar(lwd = lineSize),
                     default.units = "native")
        }

      }
      
      grid::upViewport() 
    }
  }
}