MCMCImpute: This is the flagship function of HiCImpute. It identifies...

In Queen0044/HiCImpute: HiCImpute: An R Package Implementing "Bayesian Hierarchical Model for Identifying Structural Zeros and Enhancing Single-cell Hi-C Data"

This is the flagship function of HiCImpute. It identifies structural zeros and imputes sampling zeros under a Bayesian framework. The outputs can be used to facilitate downstream analysis such as clustering, subtype discovery, or 3D structure construction.

Description

This is the flagship function of HiCImpute. It identifies structural zeros and imputes sampling zeros under a Bayesian framework. The outputs can be used to facilitate downstream analysis such as clustering, subtype discovery, or 3D structure construction.

Usage

MCMCImpute(
  scHiC,
  bulk = bulk,
  startval = c(100, 100, 10, 8, 10, 0.1, 900, 0.2, 0, replicate(dim(single)[2], 8)),
  n,
  epsilon1 = 0.5,
  epsilon2 = 5,
  mc.cores = 1,
  cutoff = 0.5,
  niter = 30000,
  burnin = 5000
)

Arguments

scHiC	The single-cell Hi-C matrix. It can take three types of formats. The preferred format is a single-cell matrix with each column being a vector of the upper triangular matrix without including the diagonal entries of the 2D matrix of a single-cell. Another types of formats are a list with each element being a 2D single-cell contact matrix, or a 3D (n\times n\times k) array that has k matrices of dimension n\times n. HiCImpute automatically transforms these two types of input into a matrix with each column being the vector of upper triangular matrix of a single-cell. For a single-cell matrix of size n \times n, the length of the vector should be n\times(n-1)/2. We only need the upper triangular matrix because the Hi-C matrix are symmetrical.
bulk	The bulk data. It can take two types of formats. A 2D bulk matrix of dimension n\times n or a vector of the upper triangular entries of 2D bulk matrix. It can provide information for priors settings. If bulk data is not available, simply set it to be NULL, and MCMCImpute will sum up the single-cells to construct a bulk data.
startval	The starting value for the vector of parameters Θ=(α, μ^γ, β, μ, a, δ, b, π, s, μ_1, \cdots, μ_K). See xie et al. for the details of these parameters. The default value is as set in the function.
n	Integer. The dimension of single-cell matrix.
epsilon1	The range size of δ that is used to monitor the prior mean of π_{ij}, the probability that the pair (i,j) do not interact. The default value of ε_1 is 0.5.
epsilon2	The range size of B that is used to monitor the prior mean of μ_{ij}, the intensity of interaction between pair (i,j). The default value of ε_2 is 5.
mc.cores	The number of cores to be used in mclapply function that can parallelly impute the matrix. The default value is 1 (no parallelization), but the users is advised to a higher number to increase computational speed if their computer has parallel computing capability.
cutoff	The threshold of π_{ij} that is used to define structural zeros. The default value is 0.5. That is, if the probability of being a SZ is greater than 0.5, then the pair (i,j) are labelled as not interacting due to underlying biological mechanism.
niter	The number of iterations for the MCMC run. Default is 30000.
burnin	The number of burn-in iteration. Default is 5000.

Value

A list of posterior mean of probability (SZ), the imputed data without defining SZ (Impute_All), and imputed data with SZ, using the threshold (Impute_SZ).

Examples

data("K562_T1_7k")
data("K562_bulk")
scHiC=K562_T1_7k
T1_7k_res=MCMCImpute(scHiC=K562_T1_7k,bulk=K562_bulk,
startval=c(100,100,10,8,10,0.1,900,0.2,0,replicate(dim(scHiC)[2],8)),n=61,mc.cores = 1,
cutoff=0.5, niter=100000,burnin=5000)

Queen0044/HiCImpute documentation built on Oct. 9, 2022, 9:30 a.m.