使用d_ply对For/while循环进行优化

Question

我正在运行一个自定义函数，它为data.frame的每一行生成并保存一些情节：

zz <- "chr  start end   name  max
chr11 66184332 66197785 NPAS4_CBP20 90
chr11  62666002 62683613 BC047540_CBP20 100   
chr1 9824542  9828548  MIR3687_CBP20  500
chr6  33239767 33259282 B3GALT4_Pol2   1000
chr20  244996112   245029580   HNRNPU-AS1_Pol2   450
chr20 62487823 62525914 ABHD16B_Pol2   370
chr12 121146198   121179996   ACADS_Pol2  90"

my.genes <- read.table(text=zz, header = TRUE)

该函数有两个步骤，一个是慢的(~40秒)，但只需要对每个chr运行一个，然后对该chr的每一行运行。在伪码中，应该是这样的：

for each chr createData
   for each nameInChr do somethingWithData.

我的问题是，如何优化这一点？嵌套d_ply？按照chr对data.frame进行排序，然后对唯一的应用(mygene$chr)进行某种形式的应用？

如果不包含实际的功能，我很抱歉，但这是一个很长的问题，我认为这是一个更好的实践/理论问题。但是，如果需要，我可以添加适当的代码。

编辑

这里是简化函数(plotGviz)。实际上，我不认为d_ply会工作，因为它会对每一行运行UcscTrack (我是正确的吗？)最终的目标是对每个唯一的染色体运行UcscTrack，然后使用这些信息为每个基因(名称)构建小区。这两个步骤(UcscTrack和绘图)都很费时，但基本原理是，首先运行代码的非唯一部分(适用于chr)，可以节省整个表的时间--该表是我的一小部分。

自定义plotGviz中的函数不能仅以我设置它们的方式进行优化。

################
### libraries ##
library(Gviz)
library(GenomicRanges)
library(GenomicFeatures)
library(data.table)
library("RColorBrewer")
#######################


d_ply(my.genes, .(chr, name), plotGviz)

plotGviz <- function(gene) {
   chr <- gene$chr
   mygene.start <- gene$start
   mygene.end <- gene$end
   mygene <- gene$name
   max.cov <- gene$max

#################################################
## this part runs only once for each unique chr
## UcscTrack is what takes most time

   ## Gene annotations ##
   knownGenes <- UcscTrack(genome=gen, chromosome=chr,
   track="knownGene",  trackType="GeneRegionTrack",
   rstarts="exonStarts", rends="exonEnds", gene="name", symbol="name",
   transcript="name", strand="strand", fill="#FF7F00", name="UCSC Genes", geneSymbols = TRUE, showId = TRUE)

   refGenes <- UcscTrack(genome=gen, chromosome=chr,
   track="refGene", trackType="GeneRegionTrack", 
   rstarts="exonStarts", rends="exonEnds", gene="name", symbol="name",
   transcript="name", strand="strand", fill="#FF7F00", name="RefSeq Genes", geneSymbols = TRUE, showId = TRUE)

   ## axis scale ##
   ideoTrack <- IdeogramTrack(genome = gen, chromosome = chr)
   # plotTracks(ideoTrack, from = mygene.start, to = mygene.end)

   axisTrack <- GenomeAxisTrack(scale=0.25)


####################################
## now for each name in unique chr
######################

   ## construct tracks ##


   ## ChIP-seq coverage from BAM ##
   bamPol2 <- "~/bioinfo/srp_chip_seq/data/reads/merged_reads_CBP20line/Pol2.bam"
   bamInput <- "~/bioinfo/srp_chip_seq/data/reads/merged_reads_CBP20line/Input.bam"

  ## histogram
   bTrackPol2 <- DataTrack(range = bamPol2, genome = gen, chromosome = chr,
      name = "Pol2", type = "histogram", col.histogram="#984EA3",
      ylim=c(0,max.cov))

   bTrackInput <- DataTrack(range = bamInput, genome = gen, chromosome = chr,
      name = "Input", type = "histogram", col.histogram="#999999",
      ylim=c(0,max.cov))


   #################
   ## plot tracks ##

   pdf(paste("./plots/", mygene,"_cov.pdf", sep="")) 

   plotTracks(list(ideoTrack, axisTrack, bTrackCBP20, pTrackCBP20, bTrackPol2, pTrackPol2, bTrackInput, refGenes, bTrackRNAcov), from = mygene.start, to = mygene.end, fontfamily="Helvetica", background.title="white", col.title="black", col.axis="black")

   plotTracks(list(ideoTrack, axisTrack, bTrackCBP20, pTrackCBP20, bTrackPol2, pTrackPol2, bTrackInput, knownGenes, bTrackRNAcov), from = mygene.start, to = mygene.end, fontfamily="Helvetica", background.title="white", col.title="black", col.axis="black")
   dev.off()

   # plotTracks(list(axisTrack, refGenes, bTrackCBP20, pTrackCBP20, bTrackPol2, pTrackPol2, bTrackInput, bTrackRNA), from = mygene.start, to = mygene.end, fontfamily="Helvetica", background.title="white", col.title="black", col.axis="black")

   # head(displayPars(bTrackPol2))
}

edit2

我的临时解决方案使用唯一chr的for循环，在子设置每个chr的所有名称之前创建UcscTrack，并为绘图创建一个d_pply。基本上是在两个地方吐出这个大的功能。

for (chrom in unique(my.genes$chr)) {
   print(paste("creating annotation for ", chrom, sep=""))

   ## Gene annotations ##
   knownGenes <- UcscTrack(genome=gen, chromosome=chrom,
   track="knownGene",  trackType="GeneRegionTrack",
   rstarts="exonStarts", rends="exonEnds", gene="name", symbol="name",
   transcript="name", strand="strand", fill="#FF7F00", name="UCSC Genes", geneSymbols = TRUE, showId = TRUE)

   refGenes <- UcscTrack(genome=gen, chromosome=chrom,
   track="refGene", trackType="GeneRegionTrack", 
   rstarts="exonStarts", rends="exonEnds", gene="name", symbol="name",
   transcript="name", strand="strand", fill="#FF7F00", name="RefSeq Genes", geneSymbols = TRUE, showId = TRUE)

   ## axis scale ##
   ideoTrack <- IdeogramTrack(genome = gen, chromosome = chrom)
   # plotTracks(ideoTrack, from = mygene.start, to = mygene.end)

   axisTrack <- GenomeAxisTrack(scale=0.25)

   ## select genes in chromosome
   df.g <- subset(my.genes, chr == chrom)
   d_ply(df.g, .(name), plotGviz)
}

Answer 1

试试data.table，它会比plyr快。下面是解决问题的一般方法：

library(data.table)
dt = data.table(my.genes)

dt[, .SD[, do_smth(), by = name], by = chr]

下面是一个不同问题中的上述方法的示例-- how to integrate properties defined on multiple rows using a data.frame or data.table long format approach