Biophysical Society Thematic Meeting| Les Houches 2019

Multiscale Modeling of Chromatin: Bridging Experiment with Theory

Poster Abstracts

12 - POS Board 12 IDENTIFICATION AND INCORPORATION OF COPY NUMBER INFORMATION FOR CORRECTING HI-C DATA OF CANCER CELLS Ahmed I.S. Khalil 1 ; Anupam Chattopadhyay 1 ; Amartya Sanyal 2 ; 1 Nanyang Technological University, School of Computer Science and Engineering, Singapore, South-West, Singapore 2 Nanyang Technological University, School of Biological Science, Singapore, South-West, Singapore Hi-C and its variant techniques have been developed to capture the genome-wide chromatin interactions. Filtering and normalization of Hi-C data are essential for accurate modelling and interpretation of genome-wide contact map. Although several methods have been developed, they usually account for genome sequence biases and ignore the data-driven biases such as copy number variations (CNVs). Cancer genomes are plagued with widespread multi-level structural aberrations of chromosomes. These pose a challenge to accurately identify CNV profiles and their subsequent usage for better filtering and normalization of contact map. Here, we propose a framework that utilize read depth (RD) of coverage from Hi-C or 3C-seq datasets to first identify their CNV profiles including both large-scale and focal events. Then, we integrate these CNV tracks with other systematic biases for correcting the interaction frequencies using Poisson regression model. We demonstrated that the RD signal, computed by combining valid read pairs with single-side mapped reads and read pairs mapped to same fragment, recapitulates the RD profile derived from input control of ChIP-seq data of same cell line. This allows identification of focal CNVs beside large-scale segmental alterations. Interestingly, we discovered that the chromatin contact frequencies of cancer cells are highly correlated with large-scale CNVs suggesting that ignoring these large segmental alterations may lead to biased interaction signal. We have applied this normalization technique on several Hi-C/3C-seq cancer datasets which effectively removes the data-driven biases.

66