train_k_caller: K-caller performance for a given set of eps and minPts

In BenjaminPlanterose/UMtools: An R-package for analysing Illumina DNA Methylation arrays at the fluorescence intensity level

K-caller performance for a given set of eps and minPts

Description

It allows the quantification of the K-caller's performance for a given dataset, training set of probes and parameters eps and minPts.

Usage

train_k_caller(M, U, training_set, minPts, eps, nThread = NULL)

Arguments

M	Methylated fluorescence mean intensity matrix (CpGs as rows, samples as columns)
U	Unmethylated fluorescence mean intensity matrix (CpGs as rows, samples as columns)
training_set	List. As an example, see data(training_set). More details in help(training_set)
minPts	dbscan parameter. Minimum numbers of points in the eps region to consider a core point; help(dbscan, dbscan) for more details.
eps	dbscan parameter. Size of the neighbourhood; help(dbscan, dbscan) for more details.
nThread	Number of CPU cores to employ

Value

List. Containing confusion matrix and macro-Precission, macro-Recall and Macro F1-score for 1-3 clusters and 1-4 clusters. Given the rarity of K=4 clusters, it is not fair to give the same weight in evaluation of the multi-class classification performance. To decide what combination of parameters to employ, we recommend to use macro F1-score for 1-3 clusters.

Examples

data(training_set)
train_k_caller(M, U, training_set, 3, 0.07)

BenjaminPlanterose/UMtools documentation built on Aug. 19, 2024, 4:54 a.m.