About comparison with Original HiC matrix

[ahmedabbas811]

Dear all,

I followed the example you give to predict GM12878 HiC contact map in a certain region of 2MB

To measure the correlation between this prediction and the original HiC matrix of the same region, I understand that I have to perform (log(ICE normalized counts + 1)) on the original 2MB size HiC matrix, and then measure the correlation directly with the predictions.

Is that true?

Thanks

[tanjimin]

Hi @ahmedabbas811, yes I think that's the case. We used log(x + 1) of the original Hi-C matrix as prediction target. See this line:

C.Origami/src/corigami/data/chromosome_dataset.py

Line 119 in a7c288e

mat = np.log(mat + 1)

When you compare with actual Hi-C, you can either perform log1p of the original Hi-C or perform exp(x) - 1 to the prediction to bring them to the same scale.

[ahmedabbas811]

Thanks for your answer.

I am asking now about comparison with DeepC

DeepC predicts the outputs in the format of Hi-C skeleton. Can I know how you got the original interactions (without skeleton format transforming) so that you can make the ICE normalization after that?

Thanks

[tanjimin]

Hi @ahmedabbas811 , check out this section "Performance comparison with previous methods" in our methods: https://www.nature.com/articles/s41587-022-01612-8#Sec11