GPU MrBayes

GPU MrBayes implements GPU-accelerated Metropolis-coupled Markov chain Monte Carlo (MC^3) Bayesian phylogenetic inference for protein sequence data sets.

Key Features:

• ta(MC)(3) GPU MrBayes V3.1 (MC^3 GPU implementation): Adapts MrBayes' Metropolis-coupled Markov chain Monte Carlo (MC^3) sampling to run on GPUs.
• 64-bit variable utilization: Enables processing of larger protein data sets by using 64-bit variables.
• Kahan summation: Applies Kahan summation to reduce floating-point arithmetic errors and improve numerical accuracy and convergence.
• Performance improvements: Provides up to a 2.5-fold speedup versus the fastest available software and up to a 90-fold speedup versus serial MrBayes.
• Multi-GPU support: Optimized for concurrent execution across multiple GPUs to increase computational throughput.

Scientific Applications:

• Bayesian phylogenetic inference on protein data sets: Performs Bayesian tree estimation and posterior sampling for protein sequence analyses.
• Large-scale protein sequence phylogenetics: Enables analysis of larger protein data sets that exceed the capacity of 32-bit variable implementations.
• Evolutionary modeling of complex relationships: Supports construction of phylogenetic trees for studies of complex evolutionary relationships in genomic data sets.

Methodology:

Adaptation of MrBayes' Metropolis-coupled MCMC (MC^3) algorithm to GPU parallel processing, combined with 64-bit variable support and Kahan summation to improve numerical accuracy.