LuxRep

LuxRep models genome-wide DNA methylation from bisulfite sequencing (BS-seq) by integrating technical replicates with varying bisulfite conversion efficiencies to improve methylation level estimation and differential methylation detection.

Key Features:

• Incorporation of technical replicates: Integrates multiple BS-seq libraries derived from the same biological sample, including those with varying bisulfite conversion efficiencies.
• Probabilistic general linear model: Employs a probabilistic framework formulated as a general linear model to jointly analyze multiple libraries per sample.
• Library-specific bisulfite conversion modeling: Explicitly models bisulfite conversion efficiency for each library when estimating methylation levels.
• Variational inference: Uses variational inference to accelerate approximate Bayesian estimation and enable whole-genome analyses.
• Improved methylation and differential detection: Enhances accuracy of methylation level estimates and detection of differentially methylated sites and regions.
• Validation on simulated and real data: Demonstrated improved precision in simulations and analyses of real DNA methylation datasets.
• Optimized data utilization: Enables inclusion of low-conversion-rate libraries that would otherwise be excluded, reducing the need for additional sequencing and associated experimental costs.

Scientific Applications:

• Genome-wide DNA methylation analysis: Estimation of methylation levels from BS-seq data at genome scale.
• Differential methylation detection: Identification of differentially methylated sites and regions between conditions or sample groups.
• Epigenetic studies using technical replicates: Integration of multiple libraries per biological sample to improve robustness of epigenetic inferences.
• Analysis of low-conversion datasets: Inclusion and analysis of BS-seq libraries with low bisulfite conversion efficiencies to maximize data retention.

Methodology:

Fits a probabilistic general linear model that integrates multiple BS-seq libraries per biological sample while modeling library-specific bisulfite conversion efficiencies and uses variational inference for approximate Bayesian estimation.