Parallel PPI Prediction Performance Study on HPC Platforms

Abstract

STRIKE is an algorithm which predicts protein-protein interactions (PPIs) and determines that proteins interact if they contain similar substrings of amino acids. Unlike other methods for PPI prediction, STRIKE is able to achieve reasonable improvement over the existing PPI prediction methods. Although its high accuracy as a PPI prediction method, STRIKE consumes a large execution time and hence it is considered to be a compute-intensive application. In this paper, we develop and implement a parallel STRIKE algorithm for high-performance computing (HPC) systems. Using a large-scale cluster, the execution time of the parallel implementation of this bioinformatics algorithm was reduced from about a week on a serial uniprocessor machine to about 16.5 h on 16 computing nodes, down to about 2 h on 128 parallel nodes. Communication overheads between nodes are thoroughly studied.

Hardware Acceleration of the STRIKE String Kernel Algorithm for Estimating Protein to Protein Interactions

“…The performance of the parallel STRIKE was evaluated [6] on multicore computers [6] and PC clusters [14] for a test set of 168 protein sequences and an 84-sequence training set with good algorithm scalability. The 128-node PC cluster performed this matching in about 2 hours compared to about a week on a single core x86 laptop.…”

“…This is exacerbated by the large number of very long protein sequences to match in protein databases. Other novel PPI methods, such as STRIKE [6], use only the information of protein sequences and have been parallelized [6,14], but have potential to be further accelerated by dedicated and customized hardware accelerators. Comparing the sensitivity and specificity of these PPI methods using the same source of data is an effective way to compare their performance.…”

“…Furthermore, the performance of the parallel STRIKE was evaluated [14] on dual-core and quad-core computers and on a 16-desktop computer heterogeneous cluster and a 128-node SUN computer cluster. On long protein sequence sets, it was shown that the execution time of the parallel implementation of STRIKE was reduced from about a week on a serial uniprocessor machine to about 16.5 hours on 16 computing nodes, down to about 2 hours on 128 parallel nodes.…”

“…Table 6 compares the average length (N1=N2=100) sequence pair match time of the weighted STRIKE accelerator, running at 1 GHz, to the time of the message-passing STRIKE software version on 16-node and 128-node computer clusters [14] which a total of 14112 performed sequence pair matches (168 sequences in testing set x 84 sequences in the training set). From Table 3, the matching of an average sequence pair with N1=N2=100 consumes 679701 clocks or 0.68 mseconds on the Weighted STRIKE accelerator.…”

“…In [6], we developed a multithreaded version of STRIKE on multi-core systems. In [14], we also developed and studied the performance of parallel STRIKE versions on multicore computers and Message Passing Interface Standard (MPI)based clusters. On long protein sequence sets, the execution time of a parallel implementation of this bioinformatics algorithm was reduced from about a week on a serial uniprocessor x86 laptop to about 2 hours on 128 parallel cluster nodes.…”