PSbEC : Parallel Spectrum-based Error Correction

PSbEC is a parallel spectrum-based error correction software. It is developed as a generic framework to parallelize any spectrum-based error correction and includes three major optimizations (1) dynamic load balancing , (2) cache optimized array layout, and (3) Use of shared memory threads to facilitate processing of large spectrum.

Though any spectrum-based error correction can be parallelized with our framework, currently we have demonstrated it by a parallel implementation of the error correction software Reptile.

Installation

Dependencies

• A modern, C++11 ready compiler such as g++ version 4.8 or higher or clang version 3.2 or higher.
• C++11 threads.
• The cmake build system (Version >= 2.8.11).
• A 64-bit operating system. Either Mac OS X or Linux are currently supported.
• MPI installation. Tested with MPICH2 and Intel MPI.

Compilation

Create a build directory. For example,

 mkdir build
 cd build

Configure and build the executable 'cao-preptile'. Continuing the example,

 cmake ..
 make

If successfully built, cao-reptile should be available in the build directory.

Usage

Input config file should be prepared similar to Reptile. Documentation for setting parameters for reptile in the config file is given at the Reptile page here .

We follow Reptile's conventions except for the read input format. Unlike Reptile, we require the read input to be in the fastq format. We also provide the script fastq-filter-v1.0.pl to filter reads with 'N' character.

We provide the following additional options related to the parallel algorithm and the optimizations.

1. RunType : Possible options 0, 1, and 2. Set to 0 to construct spectrum and run error correction, Set it to 1 to run only distributed spectrum construction, Set it to 2 to run only error correction.
2. KmerSpectrumInFile : Input kmer spectrum file when RunType is set to 2. Input must be generated using the WriteSpectrum option.
3. TileSpectrumInFile : Input tile spectrum file when RunType is set to 2. Input must be generated using the WriteSpectrum option.
4. CacheOptimizedSearch : Default 0, no cache optimized search. Set it to 1 for using cache aware layout for kmer/tile spectrums; 2 for cache oblivious layout of kmer/tile spectrums.
5. WriteSpectrum : Default 0. If set to 1, then the spectrum is written to an output file.
6. Threads : Default 1. No. of shared memory threads per node.
7. WorkDistribution : Default 0. Set to 0 for static distribution, 1 for dynamic distribution.

Citations

1. Nagakishore Jammula, Sriram Chockalingam, and Srinivas Aluru. Parallel Read Error Correction for Big Genomic Datasets. 22nd IEEE International Conference on High Performance Computing (HiPC 2015) (Accepted).
2. Ankit Shah, Sriram Chockalingam, and Srinivas Aluru. A Parallel Algorithm for Spectrum-based Short Read Error Correction. 26th IEEE International Symposium on Parallel and Distributed Processing, IPDPS 2012.