Overview / Usage

Epistasis Detection is a fundamental part of genome-wide association studies in modern bioinformatics. Exhaustive search of epistasis detection provides the most reliable way to identify accurate solutions, but it is both computationally and memory intensive task. This work proposes the use of heterogeneous computer architectures composed of multi-core CPUs and integrated GPUs to achieve high-performance and energy-efficient epistasis analysis.

Methodology / Approach

A set of traditional programming models, i.e., OpenCL and OpenMP, and novel standards like Data-Parallel C++ are used to boost the performance, power and energy-efficiency of bioinformatic applications. oneAPI tools such as Intel Advisor and Intel VTune Profiler are used to aid the development, by providing metrics on performance and improvement guides. Techniques used include parallelization using multi-core CPUs and integrated GPU, cache blocking and vectorization.

Technologies Used

8th Generation Intel Core CPU, Intel Integrated Graphics Gen9.5, OpenCL 2.0, OpenMP, Data-Parallel C++, oneAPI base toolkit, Intel Advisor, Intel vTune Profiler

Documents and Presentations

Repository

https://github.com/hiperbio/cross-dpc-episdet

Other links

• Published article: Heterogeneous CPU+iGPU Processing for Efficient Epistasis Detection

• CPU+GPU Epistasis Detection using OpenMP and OpenCL (https://github.com/hiperbio/het-cl-episdet)

• oneAPI Developer Conference 2020 Presentation: Rooflining Bioinformatics: Boosting Epistasis Detection with Cache-aware Roofline Model

Collaborators