Classical graph-based agglomerative methods employ a variety of similarity measures between nodes to partition PPI networks, but they often result in a poor clustering arrangement that contains one or a few giant core clusters with many tiny ones [1].
One more SATé run for each data set was employed on an 8-core cluster with a 24 hour limitation as well (Table S2).
In this paper we analyze the performances of some widely used, computational intensive, applications, like FFT, convolution and n-body simulation, comparing the performance of a multi-core cluster node, with or without the contribution of GPUs.
The CAD homologs found in core chlorophytes cluster with land plants (albeit with modest 74% bootstrap support).
Compute time on a 608 Intel Xeon core Linux cluster with 6 TB of DDR3 RAM and 20 TB SATA hard drives was 22.5 h.
Using this rather strict cutoff resulted in a set of 1729 core clusters, i.e. clusters with at least one protein for each of the 53 genomes in the study.
Similar results were obtained when confining the analyses to just the core clusters, i.e., the clusters with the largest numbers of members, where there was 98% (2,122 of 2,266) agreement between the two methods.
For example, it takes usually about 5 h to fit a catalog of 20,000 events on a 48-core linux cluster with Intel CPU [email protected].
Finally, the last step is responsible for the merge of core clusters and it is performed with a single MapReduce algorithm.
The analysis performed with 150 core clusters showed a total of 27 33 rearrangements required to explain the changes in the genomic architecture from the ancestral free-living relative (the most recent common ancestor with H. elongata and C. salexigens) to the three endosymbionts (fig. 8).
