DecGPU

DecGPU performs parallel and distributed error correction of high-throughput short reads to improve accuracy and scalability of de novo genome assembly workflows.

Key Features:

• Hybrid parallel programming: Combines CUDA and MPI to exploit GPU acceleration alongside distributed computing for error correction.
• CPU-based parallelism: Provides a CPU version using MPI and OpenMP for coarse- and fine-grained parallelism on multi-core processors.
• GPU-based parallelism: Employs a CUDA+MPI model with overlapping CPU and GPU computations to maximize throughput.
• Scalability: Distributed architecture enables processing of large-scale high-throughput short read datasets from next-generation sequencing.
• Performance and quality: Reported to outperform existing error correction algorithms in both correction quality and execution speed on simulated and real datasets.
• Assembler integration: Integrates with de-Bruijn-graph-based assemblers such as Velvet and ABySS to enhance assembly quality.

Scientific Applications:

• De novo genome assembly: Improves input read quality for more accurate de novo assemblies from short-read data.
• Variant detection: Reduces sequencing errors in high-throughput short reads to support more reliable variant calling.
• Comparative genomics: Enhances data quality for comparative analyses that depend on accurate short-read assemblies.
• High-throughput sequencing projects: Applicable to large-scale next-generation sequencing datasets requiring scalable error correction.

Methodology:

DecGPU's error correction algorithm executes in parallel across distributed environments using a hybrid model: a GPU-based implementation using CUDA with MPI that overlaps CPU and GPU computations, and a CPU-based implementation using MPI with OpenMP.