matuszelenak / hmm_sampler_CUDA

Sampling from MinION Reads

Introduction

The MinION DNA sequencing platform produces long reads with high error rates. One of the reasons for high error rates is that electrical signals produced by the sequencing machine need to be first translated into DNA sequences by a process called base calling, and this process is error prone. An alternative way of interpreting these signals is to generate multiple samples from a posterior sequence distributions defined by a hidden Markov model (HMM) representing the properties of the sequencing process. The goal of this project is to speed up this sampling by employing GPUs.

Installation

For a successful compilation you should have the following packages installed

• Boost for C++
• HDF5

Compile using this command in the src folder

nvcc -dc gpu_pathdecode.cu gpu_sampling.cu gpu_samplingv2.cu gpu_viterbi.cu gpu_utils.cu -arch sm_35
g++ -c HMM.cpp State.cpp LogNum.cpp main.cpp --std=c++11 -I /usr/include/hdf5/serial/ -lhdf5_serial -lboost_program_options -lpthread -lboost_log -DBOOST_LOG_DYN_LINK
nvcc -arch sm_35 -o  hmm_sampler gpu_pathdecode.o HMM.o State.o LogNum.o main.o gpu_sampling.o gpu_samplingv2.o gpu_viterbi.o gpu_utils.o -I /usr/include/hdf5/serial/ -lhdf5_serial -lboost_program_options -lpthread -lboost_log -DBOOST_LOG_DYN_LINK

Run with

./hmm_sampler

For usage details run with --help or -h flag.