Parallel Computational Protein Design.

Area of Science:

• Computational Biology
• Protein Engineering
• Bioinformatics

Background:

• Computational structure-based protein design (CSPD) aids in engineering protein functions.
• Existing methods combining dead-end elimination (DEE) and A* search guarantee optimal solutions but face computational bottlenecks.
• The A* search component can exhibit exponential time complexity, limiting scalability for large protein design problems.

Purpose of the Study:

• To accelerate the computational protein design pipeline.
• To overcome the computational limitations of traditional A* search in CSPD.
• To enhance the OSPREY program with GPU acceleration for improved performance.

Main Methods:

• Implemented a GPU-based, massively parallel A* algorithm within the OSPREY program, creating gOSPREY.
• Optimized the heuristic function computation for the A* search on GPUs.
• Integrated gOSPREY with existing rotamer pruning algorithms (iMinDEE, DEEPer) to handle flexibility.

Main Results:

• gOSPREY achieves up to four orders of magnitude speedup for large protein design cases compared to traditional A* search.
• The GPU implementation maintains optimality guarantees with minimal memory overhead.
• gOSPREY offers a bounded-memory mode for computationally intensive problems.

Conclusions:

• gOSPREY significantly enhances the efficiency of CSPD by leveraging GPU parallelization.
• The tool provides a scalable and efficient solution for complex protein design challenges.
• gOSPREY enables the design of proteins with improved functions and considers backbone and side-chain flexibility.