19. beta_selectNBest

19.1. Overview

beta_selectNBest selects CpG features associated with a categorical phenotype using scikit-learn feature-selection methods.

The input matrix must have CpGs in rows and samples in columns. CpGs containing missing values are removed before feature selection, and zero-variance CpGs are removed by default.

19.2. Scoring Methods

Method	Description
`anova`	ANOVA F-statistic for association between methylation values and group labels.
`mi`	Mutual information between methylation values and group labels.
`chisq`	Chi-square statistic. Input values must be non-negative.

19.3. Input Files

19.3.1. Beta matrix

A tab-delimited matrix with CpG IDs in the first column and samples in the remaining columns. Compressed input is supported.

Example:

CpG_ID   Sample_01   Sample_02   Sample_03   Sample_04
cg_001   0.831035    0.878022    0.794427    0.880911
cg_002   0.249544    0.209949    0.234294    0.236680
cg_003   0.845065    0.843957    0.840184    0.824286

19.3.2. Group file

A two-column CSV or TSV file containing sample IDs and categorical group labels. A header is optional.

Example:

Sample,Group
Sample_01,normal
Sample_02,normal
Sample_03,tumor
Sample_04,tumor

19.4. Usage

Basic usage:

beta_selectNBest \
    -i test_05_TwoGroup.tsv.gz \
    -g test_05_TwoGroup.grp.csv \
    -k 100 \
    -s chisq \
    -o selected

Useful options include:

-k, -c, --top, --topK – number of CpGs to select (default: 100)
-s, --score_function, --score-function – anova, mi, or chisq (default: chisq)
--random_state – random seed for mutual-information scoring
--keep_constant – retain zero-variance CpGs
-o, --out_prefix, --output – output prefix

Display all options with:

beta_selectNBest -h

19.5. Output

For output prefix selected, the command writes:

selected.selected_features.tsv – selected CpG-by-sample matrix
selected.feature_scores.tsv – scores, p-values, ranks, and selection status for all retained CpGs
selected.selected_cpgs.txt – selected CpG IDs

19.6. Example Data

Beta matrix