spike-in data set

[nclement]

Does anyone know where I can find an Illumina spike-in data set? We're trying to benchmark some programs, but haven't been able to find any available.

[Nix]

For chIP-seq?

[nclement]

Yes. For ChIP-seq.

[zee]

I've used this dataset previously to test our alignment algorithm.

http://www.bcgsc.ca/data/chipseq

There are some .wig files available as well.

[Nix]

Pseudo simulated spike-in datasets for chIP-seq

I've posted some pseudo simulated spike-in data sets to http://bioserver.hci.utah.edu/Supple...aperInfo/2008/ under Nix_EmpricalMethods. Basically, localized random reads are added to input sequencing data. Keys are provided so you can calculate TPRs and FDRs.

This is from a paper we just submitted:

Spike-in Data Set Generation
An application was developed to simulate single binding site chIP regions. It works by randomly selecting center positions from a genome. These are expanded to a maximum defined size (500bp) and then filtered to remove regions with a RepeatMasker base content of greater than 0.2 and a fraction of non GATC bases greater than 0.5. For each remaining region, random fragments are generated about each center position from 150 to 500bp in size. From each simulated fragment, each end is taken as a read and each base in the read mutated according to the published per cycle error frequency[12]. Reads are then aligned to the genome.

For the human spike-in dataset, 1000 regions with 1000 simulated chIP fragments were selected producing 2000 reads each. These were mapped to hg17 using the ELAND Extended aligner from Solexa. Only those regions with greater than 1000 mapped reads were chosen for use in generating the spike-in set. 60 groups containing 30 spike-in regions were created. For each of the groups, from 2 to 60 reads were randomly drawn from each of the 30 residing spike-in regions. These represent 900 spike-in regions containing 2 to 60 reads, 27,900 total. To create the actual spike-in datasets, Johnson et.al.’s control input data was combine, randomized and split in thirds, 1,698,713 reads each. To one of the thirds, the reads from the 900 spike-in regions were added. This represents the simulated chIP data, the other two simulated input data sets.

In a similar fashion, the mouse spike-in dataset was generated. Reads from 1000 regions with 1000 fragments were mapped to mm8. Regions with greater than 500 mapped reads were used to derive 87 groups, each with 10 regions, 870 total containing 1 to 87 randomly drawn reads, 38,280 total. To generate a larger simulated input dataset, data from Mikkelsen et. al. that showed little to no significant enrichment (ES.H3, ES.K9, ES.RPol, ESHyb.K9, MEF.K9, NP.K9, NP.K27, and NP.K36) were pooled along with their actual whole cell extract input data (ES.WCE, MEF.WCE, and NP.WCE), randomized, and split in thirds, 16,383,950 reads each.

You will need to combine the appropriate files to generate datasets that suit your purposes. There are a bunch of utilities at USeq (http://useq.sourceforge.net/ ) for manipulating the binary bar files if you'd rather go that route instead of using the txt files. Good luck! I'm glad you're going to bench mark your methods. -cheers, David

[Jason.Lu]

spikin-in data from RNA-Seq

Hi,
This is my first post. First want to say hi to everyone.
I wonder whether anyone knows of any RNA-Seq spike-in datasets publicly available (like latin-square data or golden spike from the affy platform).
Thanks,