cmFSM: a scalable CPU-MIC coordinated drug-finding tool by frequent subgraph mining

Area of Science:

• Computational biology
• Bioinformatics
• Data mining

Background:

• Frequent subgraph mining is crucial for analyzing large biological and chemical datasets, aiding in drug discovery and toxicity prediction.
• Existing methods face efficiency challenges due to computationally intensive graph isomorphism testing and large result sets, often requiring days for analysis and exceeding single-node memory capacity.

Purpose of the Study:

• To develop a parallel acceleration tool, cmFSM, for frequent subgraph mining algorithms.
• To address the computational intensity and memory constraints inherent in mining large-scale graph data.

Main Methods:

• Implemented cmFSM utilizing parallel techniques to optimize extension tasks and reduce computation time.
• Employed a multi-node strategy to overcome memory limitations.
• Applied parallel optimization across single-node (OpenMP), multi-node, and CPU-MIC collaborative levels.

Main Results:

• cmFSM demonstrates superior performance compared to state-of-the-art miners, even with limited parallel resources.
• Achieved up to a tenfold speedup over the best CPU-based single-node approaches.
• Showcased promising scalability for massive mining tasks in multi-node environments.

Conclusions:

• cmFSM offers a practical and efficient solution for frequent subgraph mining with extensive results.
• The tool significantly enhances the speed and scalability of analyzing large biological networks.
• Source code is publicly available for further research and application.