Team 1 report: assembly with Meraculous

Published by the Joint Genome Institute, part of the US Department of Energy. Meraculous was initially designed for haploid assembly, but currently supports diploid assembly as well. The advantages of this assembler include multi-threaded and parallelized computation, absence of error-correction for faster processing, paired-end short reads compatibility (e.g., Illumina), efficient and conservative traversal of subgraphs of the de Bruijn graph, selection of kmer set, production of a set of maximal linear sub-paths of the de Bruijn graph, and alignment of reads to assembly in order to identify useful read-pair information and closure of gaps. Meraculous has been used to assemble the Pichia stipitis genome, a 15.4 Mb genome, using 75 bp paired reads with 425x coverage. The resulting assembly covered 95% of the genome and had an N50 of 101 kb.

• Counts occurrences of each kmer in the data set.
• Removes kmers whose frequency are below a threshold provided by the user.
• For each kmer, counts the number of high-quality single-base extensions
• Classifies the 5' and 3' ends of each kmer as U, F, or X, corresponding to having zero, one, or multiple high-quality single-base extensions
• Stores the extensions of kmers with a classification in a hash
• Removes non-reciprocal U-U extensions between kmers (i.e. an extension where the end of one mer is marked as U but the other is marked F).
• Stores the linear subgraph of U-U extensions
• Selects kmers at random and extend outwards to produce contigs
• Aligns all reads to contigs via BLAST
• Assembles contigs into scaffolds using paired-end data
• Searches unaligned reads as potential gap-closers using mate-pair data

• The assembler relies on data with high quality in order to avoid error correction, also requires high coverage
• Initial release did not support polyploid genome assembly due to allowing for linear subgraphs of the de Bruijn graph only
• High disk space usage

• Requires an array of other scripts in other languages
• Most of high level scripts are written in perl
• Tested the program with the packaged test data and obtained contigs

• Main issue was new version of GCC and getting all the dependencies together ~16 hrs
• There was one non-standard perl module needed
• Files with carriage returns
• Some scripts contain error but they aren't hard to fix.

• Execute run_meraculous.sh scripts along with user-provided configuration file
• Configuration file contains info on where where data is and what format it comes in
• Creates a timestamped folder that includes directories containing results of each step and executables to suspend, resume, or restart the run from that step
• Thorough error-logging at each step, allowing you to check the errors that made a run fail and then resume the run after fixing the errors
• SGE-aware, handles qsub and monitoring jobs

• Straightforward to figure out what went wrong just requiring a basic understanding of Perl
• Handles all directory creation for you
• Logs are very useful

• Meraculous requires error correction and adapter removal. Trimming is unnecessary, as low quality reads are ignored during contig formation.
• High error rates bog down the assembler. Need to be removed.
• Kmer size chosen directly affects assembly quality

Meraculous requires an optimal kmer size for runs. KmerGenie is a program used to give optimal assembly kmer size by generating abundance histograms for many abundance histograms for many values of k. Here is a link that helped me understand KmerGenie: http://kmergenie.bx.psu.edu/.