Π-cyc: A Reference-free SNP Discovery Application using Parallel Graph Search

Motivation: Working with a large number of genomes simultaneously is of great interest in genetic population and comparative genomics research. Bubbles discovery in multi-genomes coloured de bruijn graph for de novo genome assembly is a problem that can be translated to cycles enumeration in graph theory. Cycle enumerations algorithms in big and complex de Bruijn graphs are time consuming. Specialised fast algorithms for efficient bubble search are needed for coloured de bruijn graph variant calling applications. In coloured de Bruijn graphs, bubble paths coverages are used in downstream variants calling analysis. Results: In this paper, we introduce a fast parallel graph search for different K-mer cycle sizes. Coloured path coverages are used for SNP prediction. The graph search method uses a combined multi-node and multi-core design to speeds up cycles enumeration. The search algorithm uses an index extracted from the raw assembly of a coloured de Bruijn graph stored in a hash table. The index is distributed across different CPU-cores, in a shared memory HPC compute node, to build undirected subgraphs then search independently and simultaneously specific cycle sizes. This same index can also be split between several HPC compute nodes to take advantage of as many CPU-cores available to the user. The local neighbourhood parallel search approach reduces the graph's complexity and facilitate cycles search of a multi-colour de Bruijn graph. The search algorithm is incorporated into $\Pi$-cyc application and tested on a number of Schizosaccharomyces Pombe genomes. Availability: $\Pi$-cyc is an open-source software available at www.github.com/2kplus2P