ICE normalization using JuicerTools

Question

[maxrhighsmith]

Hello, I am currently trying to analyze Hi-C contact matrices. I have been using the Juicer tool https://github.com/aidenlab/juicer/wiki/Data-Extraction and find it very helpful because it contains the Eigen and Pearson command line for extracting correlation matrices and A/B compartment. However, I have read in much of the literature that it is best practice to normalize Hi-C contact matrices, specifically with the popular ICE method https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3816492/.

Juicer provides a few normalization methods (KR, VC, VC_SQRT) however, it does not have an ICE normalization option. I know I could use the dump command to extract matrices, and then pass text matrices through ICE normalization, but after doing this the contact maps will no longer be stored in the aiden labs, .hic file format, and thus I won't be able to easily extract things like Pearson and eigenvalues using the juicer tools jar. Is there a way to easily integrate ICE normalization into the Aiden labs juicer pipeline?

Answer 1

[kalavattam]

ICE and KR are similar normalization methods that use a matrix-balancing strategy (see links below). Both are widely accepted and used in the field.

That said, if you must use the ICE method over the KR method, then you could consider to convert your extracted matrices to the .cool format; from there, you can use a number of well-developed and maintained libraries (e.g., cooler, cooltools, HiCExplorer, FAN-C) for ICE normalization and analyzing data. For example, there are functions to do Pearson, PCA, etc. analyses. You can also convert directly from .hic to .cool with either of these libraries: hic2cool, HiCExplorer. If you do decide to do some work in the .cool format, it is possible to revert to the .hic format with this ginteractions strategy that another person posted on Biostars.

• More on KR: Rao, Huntley et al. 2014 (see the supplementary methods)
• More on ICE: Imakaev et al. 2012
• Some more info and review of this family of methods
• A comparative analysis of normalization methods (has nice background info): Lyu et al. 2019

[sckinta]

Is there a way to extract ICE bias from balanced .cool file? I know that you can find weight associated with each bin by cooler dump --balanced -t bins --join --header cool_file.cool. How can we convert this weight to bias?

[kalavattam]

I might be mistaken, but I think 'bias' and 'weight' are synonyms in cooler/cooltools, e.g., link.