Improving the Discovering and Comparing of Three-Dimensional Protein Patterns

The discovery and comparison of three-dimensional (3D) patterns in protein structures are critical tasks in structural bioinformatics, particularly for identifying functional sites and potential drug targets. In this work, we present significant enhancements of Geomfinder, a ligand-independent algorithm for identifying and comparing 3D protein patterns. Our contribu- tions include the following key improvements: i) a mechanism that reduces the number of unnecessary site comparisons; ii) an optimization that lowers the computational complexity of partial score calculations from O(N×M) to O(N+M); and iii) an updated OpenMP-based parallel version to accelerate the site discovery and comparison processes further. Our results show significant performance improvements for a base dataset of proteins. Results show that the optimized sequential Geomfinder achieves up to 26.3x speedup, compared to the original sequential Geomfinder, for relatively large proteins (reducing from approximately 1 hour to approximately 2 and a half minutes). And our OpenMP Geomfinder proposal, combined with the sequential optimization, reaches a speedup of more than 1000x compared to the sequential original Geomfinder. This novel optimized and parallel Geomfinder version allows for processing large protein structures that were not reasonable to compare before.