cmFSM

cmFSM accelerates frequent subgraph mining to identify recurring substructures in drug molecular and biological graphs for applications such as drug identification and toxicity prediction of unknown compounds.

Key Features:

• Parallel extension distribution: Distributes extension tasks across multiple computing resources to reduce computation time in frequent subgraph mining.
• Multi-node memory distribution: Distributes data and mining results across multiple nodes to overcome single-node memory constraints when result volumes are large.
• Fine-grained OpenMP parallelization: Leverages OpenMP for intra-node parallelism to optimize CPU resource usage on single nodes.
• Multi-node multi-process acceleration: Uses multi-process distribution across nodes to manage large-scale data and computational demands.
• CPU-MIC collaborative optimization: Combines traditional CPUs with Many Integrated Core (MIC) processors to enhance parallel processing capabilities.
• Graph isomorphism handling: Addresses the computational intensity of testing graph isomorphisms inherent in frequent subgraph mining.
• Scalability and performance: Demonstrates scalability for massive multi-node mining tasks and reports up to one order of magnitude speedup over best CPU-based approaches on single nodes.

Scientific Applications:

• Drug and core-structure identification: Identifies recurring substructures in drug molecular and biological libraries to find candidate drugs or core biological structures.
• Toxicity prediction: Supports prediction of toxicity for unknown compounds via discovered frequent subgraphs.
• Biological network analysis: Enables rapid analysis of large biological networks through accelerated frequent subgraph mining.
• Large-scale mining for drug development: Facilitates massive mining tasks relevant to drug development and related research areas.

Methodology:

Distribute extension tasks across computing resources; distribute data and results across multiple nodes; implement fine-grained OpenMP parallelization on single nodes; apply multi-node multi-process parallel acceleration; perform CPU-MIC collaborative parallel optimization.