Chromosome capture experiments, like Hi-C, typically bin the genome into distinct bins. A contact matrix, $M$, is given by $M_{i,j} = |\{pairs | mate_1 \in bin_i, mate_2 \in bin_j\}|$.

If one of the read mates spans a bin boundary, where should that pair be assigned?

For example, if $mate_1$ overlaps $bin_i$ and $mate_2$ overlaps both $bin_j$ and $bin_{j+1}$, should this read pair be assigned to $M_{i,j}$ or $M_{i,j+1}$? Should it be randomly assigned to either bin with equal probability? Should it be assigned to both?

I know this may be a rare event, but if Hi-C datasets are generated using billions of reads (such as Rao et al., Cell, 2014), this is bound to happen.


1 Answer 1


[Old question, but possibly still relevant to answer.]

This will not necessarily be rare. One or the other of the mates could span the ligation junction itself, in which case your read start and end could be entirely different chromosomes. In fact, some tools specifically measure the quality of Hi-C libraries by measuring the number of such ligations directly observable within reads. This will be a much larger number of reads than bin-crossing reads.

The safest thing to do in all such cases (IMO) is to take the read start position as the location of interest for each mate. That way, if a junction does occur, you still have an unambiguous position which is your best guess of where each end of the ligation product actually originates. This gets around both the minor issue of the bin-crossing problem and the major issue of the ligation junction problem. There is probably some imprecision here and you throw out some information, but sorting out the complex possible situations otherwise seems like a lot of hassle for minimal gain.

  • 1
    $\begingroup$ I think what I was referring to here were undigested contiguous reads, like Figure 1 of the HiCUP method paper describes. In this case, HiCUP marks this as an invalid read pair and filters them out. From my experience, there weren't enough of these occurrences (thankfully) that including or excluding them would have affected the contact matrix that much. At the time I asked this question, the qc3C method you referenced wasn't published yet, but I think that's a good reference to link here, now that it is available. $\endgroup$ Jan 5 at 17:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.