suggestions to generate normalized BigWig file

[fhb]

Hi everyone,

I would like to create two BigWig files to display RNA-seq density histogram in the genome browser. My main dificulty is to have the histogram normalized taking into account that there is different number of aligned sequences between the two initial bam files. Can anyone suggest a way of doing it? Or a tool?

as of now I am doing:

coverageBed > to generage a .bedg file
bedGraphToBigWig > to generage .bw file

Can I simply correct number of reads in both files to a fixed number of total reads aligned (10 millinon for example)?

Thanks very much!

[biomattia]

Hi,
I was thinking about a similar issue. I was thinking about changing each sample according to a scaling factor, rather than reducing each data set to the same number of reads, or using the library size as reference. For example, I think Cufflinks determines the 3rd quartile of the loci FPKM. Details are not entirely clear on that but I would determine the 75th of the FPKM >1 as reported by other users. Then the coverage reported in the bedgraph can be multiplied by this scaling factor.
Alternatively I was thinking about random subsetting the bam file of the alignments before using coverageBed. The proportion of reads to be randomly pulled out should be determined again based on the scaling factor.
Eventually, I think the 1st solution (multiplying the coverage by the scaling factor) is both easier and more appropriate if you do not want to display the actual reads in the genome browser but only the coverage.
Any ideas?

[mbjohnson]

Hi! Did either of you decide on a good way to do this? I've just been trying to do the same.
Thanks!
Matt

[biznatch]

I've only done this for ChIP-Seq not RNA-seq so not sure if it applies, but I first normalize my .wig files by multiplying the coverage/height at each position by a scaling factor, as described by others above Eg. if there are 20 million reads and I want to scale to 10 million I multiply by 0.5. For ChIP-Seq I've found that most of my data has around 20 million reads so I've been scaling everything to 20 million, just for consistency, and so it doesn't change the actual data too much (although if a sample has way less reads it doesn't work well, eg. I was looking at one set of data that only had less than 5 million reads in their input sample, so scaling to 20 million doesn't look right). After scaling I then generate the bigwig files using wigToBigWig.

[mbjohnson]

Thanks, biznatch!

I've been using bedtools' genomeCoverageBed to convert bam to bedgraph, then ucsc's bedGraphToBigWig. So there's no wiggle file step, but I assume the same principle applies to the data values in the bedgraph file.