Few days back, Sbastien Boisvert, the author of highly efficient Ray assembler, commented on relative performance of Ray vs SPAdes assemblers (also read here). The SPAdes group completed their comparison and updated the benchmarking table on their website. Alexey Gurevich of SPAdes team was kind enough to explain their observations in the comment section of an earlier rectangular graph-related commentary. We request readers to post additional comments under the same thread so that they all stay at the same place and can be easily followed.
Readers are warned that the above exchanges are not meant to declare one single winner among all assemblers. Genome assembly has always been a difficult problem, and has only gotten more complex with arrival of short reads. Each team writing a new assembly program tries to optimize between several factors -
(i) type of nucleotide data - small genome, large repeat-prone genome, highly polymorphic genome, single cell, metagenome, transcriptome, etc.
(ii) type of sequencing technology - read length, read quality, paired ends, pairing distances, etc.
(iii) type of computer hardware - single CPU, multicore, computer cluster, Condor, GPU, Hadoop, amount of available RAM, etc.
With so many factors in play and with each genome assembly team learning from each other, it may be better to understand at a simple level how each assembler does and what problem it was most optimized to address that hoping to find one magic bullet. The challenges are no different in sequence alignment space, where arrival of Bowtie did not make BLAST obsolete, because two alignment programs excel in solving two different kind of problems.
Do not forget to check our new membership site with a lot more information on bioinformatics.