OpenCL Layout Demo for a Large Biological Network

This video demonstrates the OpenCL-accelerated Fruchterman-Reingold layout of BINDhuman Network (19,905 nodes, 38,706 edges).

AllegroLayout OpenCL Speedup

The following chart shows the speedups of our OpenCL implementation on the various devices against our single-threaded implementation, which is included as a fallback implementation for the legacy systems not supporting OpenCL. The GPUs are not high-end graphics so that you can get more speedup on the latest graphics hardware.

AllegroLayout vs. Prefuse Force-Directed

Prefuse is a default force-directed layout of Cytoscape and uses the Barnes–Hut approximation algorithm of which the computational cost is O(n log(n)), where n is the node count, while AllegroLayout computes all the node-node interaction forces without any approximations so that it is more accurate but has O(n²). However, our highly optimized OpenCL implementation of AllegroLayout is still much faster than the Prefuse Barnes–Hut approximation algorithm.

Test Networks