Testing SOAPdenovo2 Prelease Version – III

This commentary is a continuation of previous one. We are adding some additional information in case you want to explore the pregraph module further.

Here is the full set of input parameters for the sparse pregraph module -

USAGE:
pregraph_sparse -s readsInfoFile -K kmer -z GenomeSize[-g MaxKmerEdgeLength -d KmerFreqCutoff -e KmerEdgeFreqCutoff -r run_mode -p n_cpu] -o OutputFile
-s readsInfoFile: The file contains information of solexa reads
-K kmer : k value in kmer
-g MaxKmerEdgeLength(default 15): number of skipped intermediate kmers, support 1-25.
-z GenomeSize(required): estimated genome size
-d KmerFreqCutoff(optional): delete kmers with frequency no larger than (default 1)
-e KmerEdgeFreqCutoff(optional): delete kmers’ related edge with frequency no larger than (default 1)
-r run mode(default 0): 0 build graph & build edge and preArc, 1 load graph by prefix & build edge and preArc, 2 only build graph
-p n_cpu(default 8): number of cpu for use
-o OutputFile: prefix of output file name

Reader Rayan pointed out that the module is most likely implementing the method presented in this paper. We believe he is right based on our understanding of the communications with BGI team. With the suggested method, you can change the amount of RAM needed by adjusting the ‘-g’ parameter (number of skipped intermediate kmers). If the parameter is set as 1 (1 intermediate kmer skipped), no savings in RAM usage will be seen.

Here are two relevant snippets from the paper, describing the method and what it can achieve -

In this paper, we demonstrate that constructing a sparse assembly graph which stores only a small fraction of the observed k-mers as nodes and the links between these nodes allows the de novo assembly of even moderately-sized genomes (~500 M) on a typical laptop computer.

Specifically, instead of storing every single k-mer (in a de Bruijn graph) or read (in an overlap graph) as nodes, we store a sparse subset of these nodes while still ensuring the assembly can be performed. Here, we demonstrate that this approach greatly reduces computational memory demands without sacrificing the accuracy of assembly.

From:http://www.homolog.us/blogs/2012/07/05/testing-soapdenovo2-prelease-version-iii/

  • 本文由 整理发表
  • 网站部分文章源自互联网,若未正确标注来源,请联系管理员更新。文章转载,请务必保留本文链接

您必须才能发表评论!