Empty hicexp

[alkurowska]

I am running make_hicexp() on my data for 14 chromosomes. My spares matrices seem to be fine for each chromosome. Somehow the hicexp@hic_table is empty for last two chromosomes, but it doesn't give me any error. I run it for each chromosome separately. I have realized that, when I set filter = FALSE my hicexp object looks normal. But is it possible that all of the interactions will be filtered out? Those are also the biggest chromosomes, so they should have the highest number interactions.

Upon checking rowMeans() of hicexp@hic_table, I know that the min value is 4.62 and upon checking 0 in IF column for each sample, I have approx. 100 interactions with 0, majority found only in 1 sample. Therefore with arguments zero.p = 0.8 and A.min = 5, I should preserve most of my interactions.

[mdozmorov]

It is hard to answer without playing with the data, knowing data resolution. My guess is that the data is sparse and, especially for large chromosomes, the proportion of zeros and/or average interaction frequency may be below thresholds. The interplay among these parameters is a bit more nuanced - we have a symmetric, adjacency-like, matrix.

In general, if such cases happened, set filter = FALSE. It will keep all interactions, even if they are low or zero. Such interactions rarely will be detected as differential. The results may have slightly less significant p-values because noise in low interaction frequencies may increase variability. The workaround is to examine the results and filter them by logCPM (equivalent to A), logFC, or even D (distance), and calculate your own p.adj after such filtering to gain power. That should be data-driven.