An R implementation of the ZipHiC as described in "ZipHiC: a novel Bayesian framework to identify enriched interactions and experimental biases in Hi-C data" (Osuntoki et al., 2022). The ZipHiC incorporated the Potts model (Wu, 1982) to account for spatial dependency and to increase the number of components from the standard two components assumed by other methods to three components. The ZipHiC model sources of biases (genomic distances (D) between interacting loci, GC-content (GC), Transposable elements (TEs), and Accessibility score (Acc) of interacting loci) as a regression model in order to understand their effects. Due to the computational intractability of the normalizing constant in the Potts model, ZipHiC used a likelihood-free approach (Approximate Bayesian Computation) (Beaumont et al., 2009) to overcome this problem. Finally, the ZipHiC detects significant interactions using calculated posterior means from the Metropolis-Within-Gibbs sampler.
Regression model of the sources of bias is given as
log(λ(k)s) =β(k)0 + β(k)1log(Ds) + β(k)2log(GCs) + β(k)3log(TEs) + β(k)4log(Accs).
where λ(k)s with s={i, j}, k= 1,2,· · ·, K.
Note that this works on Linux or MacOS or Windows systems. For genome-wide analysis, the function can be run in parallel computing, by dividing into sub-matrices and specify the number of cores to use.
The run_metropolis_MCMC_betas function generates the calculated estimates of the interaction parameter in the Potts model, the posterior means and MCMC chains of each components, and the labelling information.
Wu, F.Y. (1982). The potts model. Reviews of Modern Physics, 54(1), 235. Mark A. Beaumont, Jean-Marie Cornuet, Jean-Michel Marin, Christian P. Robert. Adaptive approximate Bayesian computation. Biometrika, Volume 96, Issue 4, December 2009, Pages 983–990, https://doi.org/10.1093/biomet/asp052