What is a good ChIPseq peak distance cutoff?

Question

[NHEJ]

I heard from a lot of different sources that beyond 10 kbp, genes beyond this distance from their respective ChIP-seq peaks are no longer of biological interest (as they are too far away from their peaks to be worth the $ of biologically validating them). However, I am looking for an actual source that says that 10 kbp (or some value in this ballpark) is the "accepted" cutoff. I haven't been able to locate any publications discussing the question of this cutoff. Could someone please point me in the right direction?

[seidel]

I think there's some crucial context missing from your question. If you know a peak "belongs" to a gene, regardless of distance, the testable hypothesis is: (1) does this peak contribute to regulation of this gene? How is this ever "no longer of biological interest"? A DNA sequence can contribute to gene regulation across a vast range of distances. Perhaps, you don't actually know that the peak "belongs" to the nearest gene, and may serve as an enhancer for some un-identified gene? Either way, the question is also hard to answer because it's not clear what the assay for biological validation is in your case (hard to weigh the cost of an unknown). On the other hand, googling "enhancer gene distance" brings up a lot of reviews that might contain a distance graph to support what you're after. Given that sequences can act across a wide range of distances, especially depending on context (development, tissue, etc.), it's not clear to me what you're really after.

[Ming Tommy Tang]

An enhancer could be very far away from its target gene.

For example, enhancer of shh gene located 1Mb away. chromatin is in a 3D conformation, linear distant regions maybe close to each other. The 10 kb cut-off is used to assign peaks to the target genes, and it is arbitrary and is a simplified way to analyze the data. To assign a peak to a gene, one must combine other data like Hi-C, ChIA-PET and gene expression data.

Answer 1

[Ryan Dale]

This paper by Ghavi-Helm et al makes a similar argument -- that historically, people only look 10kb away from genes because that's where they expect to find functional regulatory sites.

But in Figure 2a of that paper, they show that 73% of significant long-distance interactions (measured via many 4C assays) are >50kb from the gene. It's unclear how many of these are actually functional, but it does show that an arbitrary 10kb threshold will miss over three quarters of physical interactions.