Overlap Death Match!

Intersect/overlap of genomic data (possibly from bed/bedGraph files) is implemented by

• data.table::foverlaps in R.
• GenomicRanges::findOverlaps in R.
• bedtools intersect command line program.

Demonstration

See demo.Rterm for the terminal output during my talk.

Which is fastest?

Recent versions of these packages are all pretty fast, see slides for details. The only big winner is data.table::fread, which is much faster than read.table or rtracklayer::import for reading big bed/bedGraph files.

Do they all give the same results?

They all give the correct results, if used correctly. The only issue is that chromStart is 0-based and chromEnd is 1-based in bedGraph files, so you need to use chromStart+1 to get correct results in R. More specifically, if you read a bed file into R as a data.frame with columns chrom, chromStart, chromEnd, you need to use IRanges(chromStart+1L, chromEnd) or data.table(chromStart=chromStart+1L, chromEnd) as input to findOverlaps/foverlaps.

How to reproduce these results?

The bedGraph files are big so I did not put them online anywhere, which makes it impossible to re-do the timings in TF.benchmark.RData.

However a subset of the data is available:

• http://cbio.ensmp.fr/~thocking/data/overlap-benchmark.tgz
• http://cbio.ensmp.fr/~thocking/data/full.strand.list.RData