Ensembler: Enabling High-Throughput Molecular Simulations at the Superfamily Scale
Fig 1
Diagrammatic representation of the stages of the Ensembler pipeline and illustrative statistics for modeling all human tyrosine kinase catalytic domains.
On the left, the various stages of the Ensembler pipeline are shown. The red labels indicate the corresponding text description provided for each stage in the Design and Implementation section. On the right, the number of viable models surviving each stage of the pipeline is shown for the 93 target TK domains and for two representative individual TK domains (SRC and ABL). Typical timings on a computer cluster (containing Intel Xeon E5-2665 2.4GHz hyperthreaded processors and NVIDIA GTX-680 or GTX-Titan GPUs) is reported to illustrate resource requirements per model for modeling the entire set of tyrosine kinases. Note that CPU-h denotes the number of hours consumed by the equivalent of a single CPU hyperthread and GPU-h on a single GPU—parallel execution via MPI reduces wall clock time nearly linearly.