GenoGAM

GenoGAM models genome-wide sequencing count data with generalized additive models to infer differential protein–DNA occupancy and other genomic signals from assays such as ChIP-Seq and DNA methylation.

Key Features:

• Generalized Additive Models (GAMs): Employs GAMs to model ChIP-Seq read count frequencies as products of smooth functions along chromosomes.
• Extension of mgcv: Implements an extension of the generalized additive models framework from the R-package 'mgcv'.
• Data Parallelism Strategy: Parallelizes computations over overlapping genomic intervals using a data-parallel strategy to scale to whole chromosomes.
• Objective Smoothing Parameter Estimation: Estimates smoothing parameters by cross-validation, removing the need for subjective binning or sliding window choices.
• Significance Testing: Performs base-level and region-level significance testing for full factorial designs to assess differential occupancies across genetic backgrounds, treatments, or combinations thereof.
• Increased Sensitivity with Error Control: Demonstrated increased sensitivity compared to existing methods while controlling the type I error rate in yeast ChIP-Seq benchmarks.
• Application Flexibility: Applicable to ChIP-Seq, DNA methylation data, and as a generic statistical modeling approach for diverse genome-wide assays.

Scientific Applications:

• ChIP-Seq differential occupancy analysis: Detects and quantifies differential protein–DNA occupancy and binding profiles across conditions and factorial experimental designs.
• Yeast ChIP-Seq benchmarking: Applied to yeast ChIP-Seq datasets where it showed increased sensitivity while maintaining type I error control.
• DNA methylation analysis: Models DNA methylation data to estimate methylation profiles and differences across samples.
• Generic genome-wide assay modeling: Serves as a statistical modeling tool for various genome-wide assays producing count or signal tracks.

Methodology:

Extends the generalized additive models framework (R-package 'mgcv') using smooth functions that model read counts as products of smooth chromosomal functions, estimates smoothing parameters by cross-validation, parallelizes computations over overlapping genomic intervals, and provides base-level and region-level significance testing for full factorial designs.