fastGenoGAM

fastGenoGAM implements genome-wide generalized additive models (GAMs) to model ChIP-Seq read count frequencies as smooth functions along chromosomes and to perform base-level and region-level significance testing.

Key Features:

• Genome-wide GAMs: Models ChIP-Seq read count frequencies as products of smooth functions along chromosomes instead of using binning or sliding windows.
• Sparsity exploitation: Leverages sparsity inherent in genomic models to reduce computational cost and memory use.
• Parameter fitting: Uses the SuperLU direct solver for parameter estimation.
• Uncertainty quantification: Computes standard errors via sparse Cholesky factorization combined with the sparse inverse subset algorithm.
• Smoothing parameter estimation: Estimates smoothing parameters objectively using cross-validation.
• On-disk storage: Employs the HDF5 library to store data on disk, minimizing memory footprint and input/output operations.
• Scalability: Targets gigabase-scale genomes (e.g., mammals) and reports processing of ~300 million parameters in under nine hours on a 60-core server.
• Performance improvement: Achieves approximately 2 to 3 orders of magnitude speedup over the predecessor implementation.
• Statistical testing: Supports base-level and region-level significance testing within full factorial designs.

Scientific Applications:

• ChIP-Seq differential occupancy analysis: Models read count frequencies and tests differential occupancy at base and region levels.
• DNA methylation analysis: Applicable to genome-wide DNA methylation data modeling and inference.
• Peak calling: Can be applied for peak detection in genome-wide assays.
• Large-genome analyses: Suited for whole-genome analyses in gigabase-scale genomes such as mammals.

Methodology:

Implements generalized additive models (GAMs) modeling read counts as smooth functions along chromosomes; exploits model sparsity; fits parameters with the SuperLU direct solver; computes standard errors via sparse Cholesky factorization and the sparse inverse subset algorithm; estimates smoothing parameters by cross-validation; stores data using HDF5; performs base-level and region-level testing within full factorial designs.