normalize_scope_group: Group-wise normalization of read depth with latent factors...

In rujinwang/SCOPE: A normalization and copy number estimation method for single-cell DNA sequencing

Group-wise normalization of read depth with latent factors using Expectation-Maximization algorithm and shared clonal memberships

Description

Fit a Poisson generalized linear model to normalize the raw read depth data from single-cell DNA sequencing, with latent factors and shared clonal memberships. Model GC content bias using an expectation-maximization algorithm, which accounts for clonal specific copy number states.

Usage

normalize_scope_group(Y_qc, gc_qc, K, norm_index, groups, T, 
                        ploidyInt, beta0, minCountQC = 20)

Arguments

Y_qc	read depth matrix after quality control
gc_qc	vector of GC content for each bin after quality control
K	Number of latent Poisson factors
norm_index	indices of normal/diploid cells using group/clone labels
groups	clonal membership labels for each cell
T	a vector of integers indicating number of CNV groups. Use BIC to select optimal number of CNV groups. If T = 1, assume all reads are from normal regions so that EM algorithm is not implemented. Otherwise, we assume there is always a CNV group of heterozygous deletion and a group of null region. The rest groups are representative of different duplication states.
ploidyInt	a vector of group-wise initialized ploidy return from initialize_ploidy_group. Users are also allowed to provide prior-knowledge ploidies as the input and to manually tune a few cells/clones that have poor fitting
beta0	a vector of initialized bin-specific biases returned from CODEX2 without latent factors
minCountQC	the minimum read coverage required for normalization and EM fitting. Defalut is 20

Value

A list with components

Yhat	A list of normalized read depth matrix with EM
alpha.hat	A list of absolute copy number matrix
fGC.hat	A list of EM estimated GC content bias matrix
beta.hat	A list of EM estimated bin-specific bias vector
g.hat	A list of estimated Poisson latent factor
h.hat	A list of estimated Poisson latent factor
AIC	AIC for model selection
BIC	BIC for model selection
RSS	RSS for model selection
K	Number of latent Poisson factors

Author(s)

Rujin Wang [e-mail]

Examples

Gini <- get_gini(Y_sim)

# first-pass CODEX2 run with no latent factors
normObj.sim <- normalize_codex2_ns_noK(Y_qc = Y_sim,
                                        gc_qc = ref_sim$gc,
                                        norm_index = which(Gini<=0.12))
Yhat.noK.sim <- normObj.sim$Yhat
beta.hat.noK.sim <- normObj.sim$beta.hat
fGC.hat.noK.sim <- normObj.sim$fGC.hat
N.sim <- normObj.sim$N

# Group-wise ploidy initialization
clones <- c("normal", "tumor1", "normal", "tumor1", "tumor1")
ploidy.sim.group <- initialize_ploidy_group(Y = Y_sim, Yhat = Yhat.noK.sim, 
                                ref = ref_sim, groups = clones)
ploidy.sim.group

normObj.scope.sim.group <- normalize_scope_group(Y_qc = Y_sim, 
                                    gc_qc = ref_sim$gc,
                                    K = 1, ploidyInt = ploidy.sim.group,
                                    norm_index = which(clones=="normal"), 
                                    groups = clones, 
                                    T = 1:5,
                                    beta0 = beta.hat.noK.sim)
Yhat.sim.group <- normObj.scope.sim.group$Yhat[[which.max(
                                    normObj.scope.sim.group$BIC)]]
fGC.hat.sim.group <- normObj.scope.sim.group$fGC.hat[[which.max(
                                    normObj.scope.sim.group$BIC)]]

rujinwang/SCOPE documentation built on Jan. 1, 2023, 5:40 a.m.