het-cl-episdet

This repository contains the implementations of exhaustive epistasis detection for second and third-order interaction searches, targeting Intel CPUs, GPUs and CPU+GPU systems. It supports single-objective and multi-objective evaluations with Bayesian K2 score and Mutual Entropy scoring functions. The CPU implementations are parallelized by using OpenMP, while GPU kernels are deployed with the OpenCL programming model.

What is Epistasis Detection?

Epistasis detection is a computationally complex bioinformatics application with significant societal impact. It is used in the search of new correlations between genetic markers, such as single-nucleotide polymorphisms (SNPs), and phenotype (e.g. a particular disease state). Finding new associations between genotype and phenotype can contribute to improved preventive care, personalized treatments and to the development of better drugs for more conditions.

Description

The repository is structured as follows:

• data_sets: Contains randomly generated input data sets. The data sets are balanced (50/50 case/controls) and contain varying amounts of SNPs and samples.
• cpu: Contains the CPU-only implementations for single-objective Bayesian K2 and Mutual Entropy scores.
• cpu_vect: Contains the CPU-only implementations using vectorization intrinsics for single-objective Bayesian K2 and Mutual Entropy scores.
• gpu: Contains the GPU-only implementations for single-objective Bayesian K2 and Mutual Entropy scores.
• scheduler: Contains the GPU implementations that receive the combinations from the CPU. Supports single and multi-objective Bayesian K2 and Mutual Entropy scores.
• heterogeneous: Contains the heterogeneous implementations for single and multi-objective Bayesian K2 and Mutual Entropy scores.

Note: The implementations contained in the heterogeneous, scheduler and gpu folders were only tested on Intel GPUs with Gen9.5 micro-architecture. Deploying these applications on other GPUs might require some changes in the code.

Setup

Requirements

• OpenCL (version 1.2 or more recent)
• OpenMP
• Intel Compiler (Not tested with other compilers)

Compilation

Compiling binaries (epi_k2 and epi_k3) in folders gpu, scheduler and heterogeneous for performing 2-way and 3-way searches using K2 Bayesian scoring:

• 2-way:

$ cd <folder>/k2_bayesian/2way && make

• 3-way:

$ cd <folder>/k2_bayesian/3way && make

Compiling binaries (epi_k2 and epi_k3) in folders gpu, scheduler and heterogeneous for performing 2-way and 3-way searches using Mutual Entropy scoring:

• 2-way:

$ cd <folder>/mutual_entropy/2way && make

• 3-way:

$ cd <folder>/mutual_entropy/3way && make

Compiling binaries (epi_k2 and epi_k3) in folders scheduler and heterogeneousfor performing 2-way and 3-way searches using multi-objective evaluation (K2 score and Mutual Entropy):

• 2-way:

$ cd <folder>/multi_objective/2way && make

• 3-way:

$ cd <folder>/multi_objective/3way && make

Compiling binary (epi) in folders cpu and cpu_vect for performing 2-way or 3-way searches using K2 score or Mutual Entropy:

• K2 score:

$ cd <folder>/k2_bayesian && make

• Mutual Entropy:

$ cd <folder>/mutual_entropy && make

Note: Depending on the OpenCL version, it might be necessary to change the header file from CL/opencl.hpp to CL/cl2.hpp.

Usage example

To run on Intel GPUs, might be necessary to disable the hangcheck with the command:

echo 0 | sudo tee /sys/module/i915/parameters/enable_hangcheck

Scheduler/Heterogeneous/GPU-only versions: Running a 3-way search with a synthetic dataset with 1000 SNPs (166,167,000 triplets of SNPs to evaluate) and 4000 samples:

$ ./epi_3way 4000 1000 

Scheduler/Heterogeneous/GPU-only versions: Running a 3-way search with a dataset in .csv format with 1000 SNPs (166,167,000 triplets of SNPs to evaluate) and 4000 samples (2000 controls and 2000 cases):

$ ./epi_3way <root>/data_sets/gen_1000SNPs_4000sam.csv

CPU-only version: Running a 3-way search with a synthetic dataset with 1000 SNPs and 4000 samples:

$ ./epi 4000 1000 3

CPU-only version: Running a 3-way search with a dataset in .csv format with 1000 SNPs (166,167,000 triplets of SNPs to evaluate) and 4000 samples (2000 controls and 2000 cases):

$ ./epi <root>/data_sets/gen_1000SNPs_4000sam.csv 3

Note: The input files containted in the folder data_sets have the SNPs displaced in columns and the samples in rows.

In papers and reports, please refer to this tool as follows

Campos R., Marques D., Santander-Jiménez S., Sousa L., Ilic A. (2020) Heterogeneous CPU+iGPU Processing for Efficient Epistasis Detection. In: Malawski M., Rzadca K. (eds) Euro-Par 2020: Parallel Processing. Euro-Par 2020. Lecture Notes in Computer Science, vol 12247. Springer, Cham. https://doi.org/10.1007/978-3-030-57675-2_38.

BibTeX:

@InProceedings{10.1007/978-3-030-57675-2_38,
author="Campos, Rafael
and Marques, Diogo
and Santander-Jim{\'e}nez, Sergio
and Sousa, Leonel
and Ilic, Aleksandar",
editor="Malawski, Maciej
and Rzadca, Krzysztof",
title="Heterogeneous CPU+iGPU Processing for Efficient Epistasis Detection",
booktitle="Euro-Par 2020: Parallel Processing",
year="2020",
publisher="Springer International Publishing",
address="Cham",
pages="613--628",
isbn="978-3-030-57675-2"
}