ClusProc: CNV clustering Procedure

In PedCNV: An implementation for association analysis with CNV data.

Description

This function chooses the optimal number of clusters and provides the assignments of each individuals under the optimum clustering number.

Usage

  ClusProc(signal, N = 2:6,
    varSelection = c("PC1", "RAW", "PC.9", "MEAN"),
    threshold = 1e-05, itermax = 8, adjust = TRUE,
    thresMAF = 0.01, scale = FALSE, thresSil = 0.01)

Arguments

signal	The matrix of intensity measurements. The row names must be consistent with the Individual ID in fam file.
N	Number of clusters one wants to fit to the data. N needs to be larger than 1 and if it is 1, error will be returned. The default value 2,3,...,6 will be used if it is missing.
varSelection	Factor. For specifying how to handle the intensity values. It must take value on 'RAW', 'PC.9', 'PC1'and 'MEAN'. If the value is 'RAW', then the raw intensity value will be used. If it is 'PC.9', then the first several PCA scores which account for 90% of all the variance will be used. If the value is 'PC1', then the first PCA scores will be used. If the value is 'MEAN', the mean of all the probes will be used. The default method is 'PC1'.
threshold	Optional number of convergence threshold. The iteration stops if the absolute difference of log likelihood between successive iterations is less than it. The default threshold 1e-05 will be used if it's missing.
itermax	Optional. The iteration stops if the time of iteration is large than this value. The default number 8 will be used if it's missing.
adjust	Logicals, If TRUE (default), the result will be adjusted by the silhouette score. See details.
thresMAF	The minor allele frequency threshold.
thresSil	The abandon threshold. The individual whose silhouette score is smaller than this value will be abandoned.
scale	Logicals. If TRUE, the signal will be scale by using sample mean and sample variance by columns before further data-processing.

Details

• adjustIf adjust is TRUE, the result will be adjusted by the silhouette score in the following criterion. For each individual, the silhouette scores are calculated for each group. The individual will assigned forcefully to the group which maximize the silhouette scores.

Value

It returns object of class 'clust'. 'clust' is a list containing following components:

clusNum	The optimal number of clusters among give parameter N.
silWidth	Silhouette related results.

Author(s)

Meiling Liu

Examples

# Fit the data under the given clustering numbers
clus.fit <- ClusProc(signal=signal,N=2:6,varSelection='PC.9')

Example output

Loading required package: Rcpp
Loading required package: RcppArmadillo
Loading required package: ggplot2
The first 5 principal components are used.
The logliklihood for signal model is -1663.629 when clustering number is 2.
The logliklihood for signal model is -1477.954 when clustering number is 3.
The logliklihood for signal model is -1395.767 when clustering number is 4.
The logliklihood for signal model is -1334.761 when clustering number is 5.
The logliklihood for signal model is -1289.571 when clustering number is 6.

PedCNV documentation built on May 2, 2019, 8:17 a.m.