Efficiently predicting large-scale protein-protein interactions using MapReduce

With a rapid development of high-throughput genomic technologies, a vast amount of protein-protein interactions (PPIs) data has been generated for difference species. However, such set of PPIs is rather small when compared with all possible PPIs. Hence, there is a necessity to specifically develop computational algorithms for large-scale PPI prediction. In response to this need, we propose a parallel algorithm, namely pVLASPD, to perform the prediction task in a distributed manner. In particular, pVLASPD was modified based on the VLASPD algorithm for the purpose of improving the efficiency of VLASPD while maintaining a comparable effectiveness. To do so, we first analyzed VLASPD step by step to identify the places that caused the bottlenecks of efficiency. After that, pVLASPD was developed by parallelizing those inefficient places with the framework of MapReduce. The extensive experimental results demonstrate the promising performance of pVLASPD when applied to prediction of large-scale PPIs.