Problem about hicPCA

[QianzhaoJ]

For calling compartment ,I tried different methods including hicPCA、HiTC、cooltools.
But I found the results of calling compartment using these methods are different . The results of HiTC and cooltools are similar, while the results of hicPCA is different in some regions of chromosomes especially chr1 and chr2 .
I don't know which should to be used for subsequent analysis . Can you give me some advices ?
hicPCA
hicPCA --matrix $h5 --o EIG1.bedGraph EIG2.bedGraph --numberOfEigenvectors 2 --geneTrack $gene
HiTC and cooltools both used default parameters.

[joachimwolff]

Hi,

can you tell me what 'is different' means in detail? Maybe a plot of the region would be good to understand your issue (sth like this). Are the PCA values just flipped? Does the pattern match the checkerboard pattern of the Pearson correlation matrix (like this)? How does our PCA and the result of HiTC and cooltools correlate to known protein associated with A or B compartment (E.g. CTCF and Cohesin with A compartment and H3K9me3 with B compartment)?

In general: It depends on the organism and the way the A / B compartments are computed. We use the method proposed by Lieberman-Aiden in 2009. For organisms like Drosophila we recommend to switch the obs/exp computation to 'norm'.

Best,

Joachim

[QianzhaoJ]

Thanks for your quick reply !
the hicPCA looks like

the HiTC looks like

They are both flipped by same bedGraph file describing the genes position .
Best
Qianzhao

[joachimwolff]

Hi,

can you plot these in the same image and make sure they represent the same region? You can use hicPlotTADs or pyGenomeTracks for this. I am not sure if the scaling is equal and I need to have a CTCF track next to it the give useful feedback.

Best,

Joachim

[LeilyR]

I would also recommend you to use --norm which takes the affect of proximity ligation assay into account, and compare the output.

[QianzhaoJ]

Hi
Thanks for your advices. I have plotted these in one image and I tries to add '--norm' ,but the problem remains .
In chr1 or chr2 ,The results of HiTC and cooltools are similar, while the results of hicPCA is different .

In some chromosomes (like chr10) , the results of three methods are similar

I will add the track of ATAC or other known protein associated with A or B compartment tomorrow.
Best
Qianzhao

[joachimwolff]

Sometimes it happens that the first eigenvector shows not A/B compartments but the chromosome arms. Please consider the second eigenvector from the PCA for chr1 and chr2. For this reason we have set the default value of --numberOfEigenvectors to 2. Given your cli call you have this eigenvector already computed.

[gtrichard]

Combining different eigenvectors for different chromosomes might not be the good solution.

The simplest would be to discard chromosome arms regions with hicAdjustMatrix. Otherwise you could call compartments on the beginning of chr1 and then on the end of chr1 separately.

But still, I don't get why this issue is hicPCA specific.

[QianzhaoJ]

Hi :
The results of hicPCA are becoming similar with the other two methods after I remove the regions of centromere using hicAdjustMatrix .

So,maybe hicPCA could improve on that .

[kayakNIH]

Hi I wonder what is the input? normalized and corrected data? I find one arm a and the arm b.