Utility Functions for DNA Methylation Analysis

This vignette demonstrates DNA methylation data preprocessing using Rtoolset functions.

Overview

DNA methylation data is typically represented as beta values (ranging from 0 to 1), but many statistical analyses require M values (logit-transformed beta values) for better statistical properties.

The beta2M() function in Rtoolset provides a simple and efficient way to convert beta values to M values, handling edge cases (0 and 1) and preserving data structure.

Beta to M Value Conversion

Basic Usage

The beta2M() function converts beta values to M values:

library(Rtoolset)

# Single value
beta2M(0.5)
#> [1] 0

# Vector
beta_values <- c(0.1, 0.5, 0.9, 0.01, 0.99)
M_values <- beta2M(beta_values)
M_values
#> [1] -3.169912  0.000000  3.169912 -6.629214  6.629214

Matrix or Data Frame

# Matrix input
beta_matrix <- matrix(
  runif(100, 0, 1),
  nrow = 10,
  ncol = 10
)
M_matrix <- beta2M(beta_matrix)

# Data frame input
beta_df <- data.frame(
  Sample1 = runif(10, 0, 1),
  Sample2 = runif(10, 0, 1),
  Sample3 = runif(10, 0, 1)
)
M_df <- beta2M(beta_df)

Custom Alpha Parameter

The alpha parameter prevents log(0) when beta values are exactly 0 or 1:

# Default alpha (1e-6)
beta2M(c(0, 0.5, 1))
#> [1] -19.93157   0.00000  19.93157

# Custom alpha
beta2M(c(0, 0.5, 1), alpha = 1e-8)
#> [1] -26.57542   0.00000  26.57542

Why Convert to M Values?

M values have several advantages for statistical analysis:

1. Better distribution: M values are approximately normally distributed
2. Improved variance: Variance is more stable across the range
3. Better for linear models: More suitable for regression and differential analysis

Complete Workflow Example

library(Rtoolset)

# 1. Load beta values (e.g., from Illumina array)
beta_data <- read.csv("methylation_beta_values.csv", row.names = 1)

# 2. Convert to M values
M_data <- beta2M(beta_data)

# 3. Continue with downstream analysis
# - Differential methylation analysis
# - Clustering
# - Visualization
# - etc.

Use Cases

1. Illumina Array Data

Convert beta values from Illumina methylation arrays:

# Load Illumina array data (beta values)
beta_data <- read.csv("illumina_beta_values.csv", row.names = 1)

# Convert to M values for analysis
M_data <- beta2M(beta_data)

# Check dimensions and structure are preserved
dim(beta_data)
dim(M_data)
colnames(M_data)  # Column names preserved

2. Differential Methylation Analysis

Prepare data for differential analysis with limma:

library(limma)

# Convert to M values
M_data <- beta2M(beta_data)

# Create design matrix
design <- model.matrix(~ group)

# Fit linear model
fit <- lmFit(M_data, design)
fit <- eBayes(fit)

# Get top differentially methylated sites
top_sites <- topTable(fit, coef = 2, number = 100)

3. Quality Control

Check for problematic beta values before conversion:

# Check for extreme values
extreme_beta <- sum(beta_data == 0 | beta_data == 1, na.rm = TRUE)
cat("Number of extreme beta values (0 or 1):", extreme_beta, "\n")

# Convert with appropriate alpha
M_data <- beta2M(beta_data, alpha = 1e-6)

Best Practices

1. Check for extreme values: Beta values of exactly 0 or 1 may indicate technical issues
2. Alpha parameter: Use default (1e-6) unless you have specific requirements
3. Data structure: Function preserves matrix/data.frame structure and names
4. Memory: For very large datasets, consider processing in chunks
5. Reproducibility: Always document the alpha parameter used in your analysis
6. Performance: Caching is enabled for faster vignette rebuilds

Technical Details

Conversion Formula

The conversion from beta to M values uses the logit transformation:

$$M = \log_2\left(\frac{\beta + \alpha}{1 - \beta + \alpha}\right)$$

where: - $\beta$ is the beta value (0 to 1) - $\alpha$ is a small constant (default: 1e-6) to prevent log(0) - $M$ is the resulting M value

Why M Values?

1. Normal Distribution: M values are approximately normally distributed, making them suitable for parametric tests
2. Stable Variance: Variance is more stable across the methylation range
3. Linear Models: Better suited for linear regression and differential analysis
4. Symmetric Scale: M values range from $-\infty$ to $+\infty$, centered around 0

Integration with Other Packages

M values work well with:

• limma: For differential methylation analysis
• minfi: For Illumina array preprocessing
• ChAMP: For comprehensive methylation analysis
• DSS: For differential analysis of bisulfite sequencing data

# Example with limma
library(limma)

# Convert to M values
M_data <- beta2M(beta_data)

# Design matrix
design <- model.matrix(~ group)

# Fit linear model
fit <- lmFit(M_data, design)
fit <- eBayes(fit)

# Get differentially methylated sites
top_genes <- topTable(fit, coef = 2, number = 100)

See Also

• Function reference: ?beta2M
• Full documentation website