MrBayes tgMC³++: A High Performance and Resource-Efficient GPU-Oriented Phylogenetic Analysis Method.

Area of Science:

• Computational Biology
• Phylogenetics
• Bioinformatics

Background:

• MrBayes is a key tool for phylogenetic inference using Bayesian statistics.
• Current GPU optimizations for MrBayes face bottlenecks, limiting likelihood estimation speed.
• Existing methods struggle with extensive evolutionary models and gamma categories.

Purpose of the Study:

• To develop a high-performance, resource-efficient method for GPU-accelerated phylogenetic likelihood estimations.
• To overcome limitations in current GPU-oriented MrBayes implementations.
• To enhance the capabilities of GPU-based phylogenetic analyses.

Main Methods:

• Proposed a novel decomposition storage model for implicit high-performance data transfers.
• Implemented a GPU-oriented parallelization strategy for likelihood estimations.
• Developed the tgMC3++ method, an optimized version of MrBayes.

Main Results:

• Achieved speedup factors of up to 178 on simulated datasets using four Tesla K40 GPUs.
• Outperformed existing GPU-oriented MrBayes methods (tgMC3, nMC3, oMC3) by factors of 1.6, 1.9, and 2.9, respectively.
• tgMC3++ supports a wider range of evolutionary models and gamma categories.

Conclusions:

• The tgMC3++ method offers significant performance improvements for phylogenetic inference on GPUs.
• This approach enhances computational efficiency and expands model support in GPU-accelerated phylogenetic analyses.
• tgMC3++ represents a substantial advancement for large-scale evolutionary studies.