CMU-SAFARI / optimal-seed-solver

Optimal Seed Solver (OSS) is a dynamic-programming algorithm that finds the optimal seeds of a read, which renders the minimum total seed frequency. It is described by Xin et al. at http://arxiv.org/pdf/1506.08235v1.pdf.

optimal-seed-solver

Background:

Optimal Seed Solver (OSS) is a dynamic-programming algorithm that finds the optimal seeds of a read, which renders the minimum total seed frequency.

Seed selection is an important step for pigeonhole based seed-and-extend read mappers. In seed selection, a read is broken into multiple non-overlapping substrings called seeds. Seeds are used as anchors to index into the reference genome. To tolerate e errors, a read is typically broken into e+1 seeds such that by the pigeonhole principle, at least one seed is error free.

The overall frequency of the selected seeds has a direct impact on the performance of the mapper. If seeds are frequent, the mapper has to verify many potential mappings through edit-distance calculation, which is a time-consuming process.

OSS aims to find the least set of e+1 non-overlapping seeds from a read using dynamic programing method. For details about the algorithm of OSS, please consult our manuscript online at: http://arxiv.org/abs/1506.08235

Code files:

This repository contains many code files. Besides hosting the OSS code, it also contains other seed selection implementations, code files for main classes and misc code files. Below we summarize each code file.

Seed selection mechanisms:

• optimalSolverLN.h, optimalSolverLN.cc: contains the code of OSS.
• optimalSolver.h, optimalSolver.cc: contains the code of an optimal solver that has quadruple-complexity. We used this code to verify the integrity of OSS.
• basicSolver.h, basicSolver.cc: contains the code of a basic solver, which selects seeds at fix positions with fixed lengths.
• fastHASHSolver.h, fastHASHSolver.cc: contains the code simulating the fastHASH seed selection mechanism.
• hobbesSolver.h, hobbesSolver.cc: contains the code simulating the hobbes seed selection mechanism.
• spacedSeedSolver.h, spacedSeedSolver.cc: contains the code simulating the spaced seed selection mechanism.
• thresholdSolver.h, thresholdSolver.cc: contains the code simulating the Gem mapper's seed selection mechanism.

MISC code files:

• KmerHash.h, KmerHash.cc: a hash function that transforms a ASCII string into a 64-bit int number.
• HashTree.h, HashTree.cc: a suffix tree to index the reference genome. Notice that this implementation is not optimized for performance. It was initially built for gathering statistics of the reference genome. For human reference genome it needs 350 GB main memory for indexing. Please consider replacing HashTree with BWT implementations if you plan to integrate OSS into a production mapper.
• RefDB.h, RefDB.cc: a database that caches the reference genome.

Main classes:

Each seed selection mechanism has a complementary main-code file. They are named in a fashion of "test(mechanism).cc".

How to run:

There are two steps to run OSS (and similarly other seed selection mechanisms):

1. Index the reference genome: $ ./testHashTree referenceName.fasta treeFile.tree

2. Run OSS: $ ./testOptimalSeedLN treeFile.tree readFile.fastq