Table 1. A summary of existing methods for Hi-C analysis.

Topics	Methods	Features	Pros	Cons	References
Bias reduction	Yaffe and Tanay	Non-parametric bias correction	Effective bias reduction	Computationally intensive, difficult interpretation	[73]
	HiCNorm	Parametric bias correction	Computationally efficient, easy interpretation	Rely on parametric assumption	[87]
	ICE	Normalization method	No assumption on any specific systemic biases	Limited for equal-sized genome partition	[74]
	SCN	Normalization method	Effective removal of DNA circularization bias	Limited for equal-sized genome partition	[86]
Genome partition	PCA	Chromosome-wide variance decomposition	Discovery of two compartments based on spatial proximity	Low resolution (several MBs) genome partition	[34]
	Dixon et al.	Hidden Markov model	Discovery of topological domain based on local chromatin interactions	Two-step procedure of bias reduction and genome partition	[45]
	Sexton el al.	Local distance-scaling model	Combining Yaffe and Tanay's bias correction model [73] with genome partition	Computationally intensive	[47]
	Hou et al.	Poisson mixture model	Model intra-domain and inter-domain interactions via two Poisson distributions	Two-step procedure of bias reduction and genome partition	[48]
	GeSICA	Markov clustering algorithm	Exploration of hierarchical sub-domain structures	Lack of bias reduction	[93]
Polymer model	Fractal globule model	Model chromatin as a knot-free configuration	NA*	NA*	[98,99]
	Equilibrium globule model	Model chromatin as a highly knotted configuration	NA*	NA*	[100]
	Random loop model	Looping is formed by random interaction between monomers	NA*	NA*	[42]
	Dynamic loop	Looping is formed by diffusional motion of monomers	NA*	NA*	[43]
	Strings and binds switch	Looping is affected by concentrations of binding molecules	NA*	NA*	[44]
Inferring consensus 3D chromosomal structure	Optimization-based methods	Optimize a target function to measure the fitting of a 3D model	Use biophysical constraints in 3D model reconstruction	Local modes, fail to model experimental uncertainties	[33,38,40]
	MCMC5C	Gaussian model	First statistical model for Hi-C experimental uncertainties	No bias removal, Gaussian variance estimate is ad hoc	[39]
	BACH	Poisson model	Combing bias removal with 3D model reconstruction	Lack of biophysical constraints, computational intensive	[103]
Evaluating structural variation of chromatin via statistical model	Optimization-based methods	Use multiple parallel runs to measure chromatin structural variation	Explore all possible 3D models optimizing the target function	Computationally intensive, sensitive to initialization	[33,38,40]
	Kalhor et al.	Population-based approach	Direct link Hi-C data to presence or absence of chromatin interaction	Fail to model experimental uncertainties	[37]
	BACH-MIX	Poisson mixture model	Combing bias removal with evaluating chromatin structural variation	Limited for studying local chromatin structural variation	[103]

^*A comprehensive comparison of pros and cons of each polymer model, which can be found in Ref. [94], is beyond the scope of this paper.