GENBoostGPU

GPU-accelerated local genetic variance of DNA methylation. GENBoostGPU computes the Module 02 local-genetic-variance features and relative local-SNP-contribution score (nested out-of-fold elastic net with an R-faithful glmnet port, Haseman-Elston, BSLMM orchestration) and Module 03 out-of-fold prediction for CpG and CpH regions or sites, on GPU or CPU. See Local genetic variance engine (Module 02). The original boosting elastic net remains available as a legacy API.

Key features

  • Adaptive window orchestration – distribute genboostgpu.orchestration jobs across one or many GPUs with auto-tuned max_in_flight concurrency.

  • Automated SNP curation – zero-variance filtering, missing data imputation, and LD clumping in genboostgpu.snp_processing.

  • Elastic net boosting core – reproducible variance decomposition and ridge refits from genboostgpu.enet_boosting.

  • Flexible I/O – load PLINK data, CuPy arrays, or parquet outputs with genboostgpu.data_io.

  • Tuning toolbox – global and per-window hyperparameter utilities in genboostgpu.tuning, including cohort-wide Optuna refits.

  • Reproducibility guardrails – documented seeding, metadata capture, and structured logging patterns for consistent reruns.

Supported platforms

GENBoostGPU targets Linux with NVIDIA GPUs (Ampere or newer) and CUDA 12.x. Multi-GPU orchestration requires RAPIDS cudf/cuML 26.2 and dask-cuda 26.2 or newer. Development and documentation can be performed on CPU-only machines by installing the mock/documentation requirements.

Get started

  • Quick start – minimal pipeline example with saved outputs.

  • Installation – environment setup for CPU docs versus GPU production.

  • User guide – deep dives on data formats, workflow, tuning, scaling, and reproducibility.

  • Tutorials – walkthroughs based on the scripts in examples/.

  • API Reference – autogenerated API reference.

  • Troubleshooting – common fixes for CUDA, RAPIDS, and Dask issues.

  • Contributing – guidelines for development, style, and tests.

  • Changelog – highlights from each release.

Contents

Citation

If you use GENBoostGPU in academic or industrial work, please cite:

Alexis Bennett and Kynon J.M. Benjamin. GENBoostGPU: GPU-accelerated elastic net boosting for large-scale epigenomics. DOI: 10.5281/zenodo.17238798.

Indices and tables