fast_protein_cluster: parallel and optimized clustering of large-scale protein modeling data

Area of Science:

• Computational biology
• Bioinformatics
• Structural biology

Background:

• The Nutritious Rice for the World project generated extensive protein model datasets.
• Clustering large sets of protein models is computationally intensive.
• Existing tools lack the speed and efficiency for massive datasets.

Purpose of the Study:

• To develop a fast, parallel, and memory-efficient package for clustering protein models.
• To improve the speed and accuracy of protein model clustering algorithms.
• To support both Root Mean Square Deviation (RMSD) and Template Modeling score (TM-score) metrics.

Main Methods:

• Implemented in C++ with OpenMP for multi-threading.
• Utilized custom streaming Single Instruction Multiple Data (SIMD) and advanced vector extension intrinsics for CPU acceleration.
• Incorporated OpenCL kernels for support of AMD and Nvidia Graphics Processing Units (GPUs).

Main Results:

• Achieved 60x speedup over qcprot and 3x over existing GPU implementations for RMSD calculations on a laptop CPU.
• New GPU code further accelerated RMSD and TM-score computations.
• Novel k-means and hierarchical clustering methods demonstrated up to 250x and 2000x speed improvements over Clusco, respectively.
• Identified significantly more accurate models compared to Spicker and Clusco.

Conclusions:

• fast_protein_cluster offers a substantial advancement in the speed and efficiency of protein model clustering.
• The toolkit provides accurate and fast clustering for large datasets, benefiting structural biology research.
• Its optimized algorithms and GPU support make it a valuable resource for bioinformatics.