Upsampling BAMs, or downsampling

Question

[HI]

I have four ChIP-seq sample, which has about 10-20M reads , I want to do downsampling or upsampling to make them have the same genome coverage, for example, 15M reads, I know the software picard can do downsampling .My question is, is there any software to do up sampling, Thank you very much!

[HI]

Thank you very much! The reason that I do downsampling or upsampling is to normalize the coverage between materials, then I can compare them with the result from the deeptools(bamCoverage normalized with RPKM in a bin) (The materials were mapped to different reference sequence，so they could not be compared with existing software), is it right to do so? The reason that I want to do upsampling is there exists a sample with quite little reads.

[i.sudbery]

The downsampling is a valid thing to do, and useful in many cases where normalision to read depth doesn't quite cut it. But upsampling is never valid, you are inventing data that doesn't really exist. Either downsample to the lowest sample, or discard the lowest sample.

Generally we pick a threshold, discard samples with fewer reads than that threshold and downsample the rest to it.

Answer 1

[Devon Ryan]

Up-sampling would just introduce noise, simply down-sample and call it done.

[ATpoint]

Adding on this, naive upsampling would invent information which is not there, while subsampling reduces existing information. Therefore downsampling is imho preferred. Note that in ChIP-seq because of immunoprecipitation efficiency and/or antibody quality the total read count is not necessarily informative. What is your analysis goal?