high performance computing on graphics processing units: hgpu.org

BACKGROUND: Techniques for reconstruction of biological networks which are based on perturbation experimentsoften predict direct interactions between nodes that do not exist. Transitive reduction removes suchrelations if they can be explained by an indirect path of in influences. The existing algorithms fortransitive reduction are sequential and might suffer from too long run times for large networks. Theyalso exhibit the anomaly that some existing direct interactions are also removed. RESULTS: We develop efficient scalable parallel algorithms for transitive reduction on general purpose graphicsprocessing units for both standard (unweighted) and weighted graphs. Edge weights are regarded asuncertainties of interactions. A direct interaction is removed only if there exists an indirectinteraction path between the same nodes which is strictly more certain than the direct one. This is arefinement of the removal condition for the unweighted graphs and avoids to a great extent theerroneous elimination of direct edges. CONCLUSIONS: Parallel implementations of these algorithms can achieve speed-ups of two orders of magnitudecompared to their sequential counterparts. Our experiments show that: i) taking into account theedge weights improves the reconstruction quality compared to the unweighted case; ii) it isadvantageous not to distinguish between positive and negative interactions since this lowers thecomplexity of the algorithms from NP-complete to polynomial without loss of quality.