[R] Needing a better solution to a lookup problem.

[Davis, Brian]

Thanks for the idea.  I think this example works fast mainly due to the limited number of matches.  For each fdf1$chr there are only 2 potential matches in fdf2.  In reality there are only 24 possible values for chr (1-22 and X, Y).  When I replace the chr seq with more realistic values, I run out of memory.  I'll try it out our server and let you know how it goes.  Fast and memory intensive will get me over the hump for now.

fdf1 <- data.frame(chr=sort(sample(seq(1:24), 100000, replace=TRUE)),p=runif(100000),d=sample(100000))
fdf2 <- data.frame(chr=sort(sample(seq(1:24), 200000, replace=TRUE)),s=runif(200000),t=runif(200000))
system.time(with(FDF <- merge(fdf2,fdf1),FDF[s>=p & p >= t,]))

Thanks again,

Brian

-----Original Message-----
From: ilaik9 at gmail.com [mailto:ilaik9 at gmail.com] On Behalf Of ilai
Sent: Wednesday, March 14, 2012 3:26 PM
To: Davis, Brian
Cc: r-help at R-project.org
Subject: Re: [R] Needing a better solution to a lookup problem.

You could try doing it without a loop (.C or other):

 (rgnsnp <- merge(region,snps))
 (rgnsnp[with(rgnsnp,STOP>=POS & POS >= START),])

Here is my test for merge+search on 100k/200k:

fdf1 <- data.frame(chr=1:100000,p=runif(100000),d=sample(100000))
fdf2 <- data.frame(chr=rep(1:100000,2),s=runif(200000),t=runif(200000))
system.time(with(FDF <- merge(fdf2,fdf1),FDF[s>=p & p >= t,]))
   user  system elapsed
  2.560   0.152   2.905

 Hope this helps
Elai

On Wed, Mar 14, 2012 at 1:27 PM, Davis, Brian <Brian.Davis at uth.tmc.edu> wrote:
> I have a solution (actually a few) to this problem, but none are computationally efficient enough to be useful.  I'm hoping someone can enlighten me to a better solution.
>
> I have data frame of chromosome/position pairs (along with other data for the location).  For each pair I need to determine if it is with in a given data frame of ranges.  I need to keep only the pairs that are within any of the ranges for further processing.
>
> Example:
> snps<-NULL
> snps$CHR<-c("1","2","2","3","X")
> snps$POS<-as.integer(c(295,640,670,100,1100))
> snps$DAT<-seq(1:length(snps$CHR))
> snps<-as.data.frame(snps, stringsAsFactors=FALSE)
>
>  snps
>  CHR  POS DAT
> 1   1  295   1
> 2   2  640   2
> 3   2  670   3
> 4   3  100   4
> 5   X 1100   5
>
> region<-NULL
> region$CHR<-c("1","1","2","2","2","X")
> region$START<-as.integer(c(10,210,430,650,810,1090))
> region$STOP<-as.integer(c(100,350,630,675,850,1111))
> region<-as.data.frame(region, stringsAsFactors=FALSE)
>
> region
>  CHR START STOP
> 1   1    10  100
> 2   1   210  350
> 3   2   430  630
> 4   2   650  675
> 5   2   810  850
> 6   X  1090 1111
>
>
> The result I need would look like
>
> Res
>
>  CHR  POS DAT
>   1  295   1
>   2  670   3
>   X 1100   5
>
>
> I have a solution that works reasonably well on small sets, but my current data set is ~100K snp entries, and my regions table has ~200K entries. I have ~1500 files to go through
>
> I haven't found a good way to efficiently solve this problem.  I've tried various versions of mapply/lapply, for loops, etc which get the answer for small sets but takes hours (per file) on my real data.  Bioconductor seemed like the obvious place to look, but my GoogleFu must not be that great.  I never found anything relevant.
>
> Any ideas or points to the right direction would be greatly appreciated.
>
>
>
> Brian Davis
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.