Novel hybrid GPU-CPU implementation of parallelized Monte Carlo parametric expectation maximization estimation method...

Area of Science:

• Pharmacokinetics and Pharmacodynamics
• Computational Science
• Biotechnology

Background:

• Population PK/PD model development is time- and labor-intensive.
• Graphical Processing Unit (GPU) technology can accelerate scientific computations.
• Accelerating PK/PD analysis is crucial for efficient model-based drug development.

Purpose of the Study:

• To develop a hybrid GPU-CPU implementation of the Monte Carlo Parametric Expectation Maximization (MCPEM) algorithm for population PK data analysis.
• To assess the performance of the hybrid algorithm compared to a CPU-only implementation.
• To evaluate the potential of GPU computing in speeding up population PK/PD analysis.

Main Methods:

• Developed a hybrid GPU-CPU implementation (MCPEMGPU) and a CPU-only implementation (MCPEMCPU) of the MCPEM algorithm in MATLAB.
• Utilized a computer with dual Xeon CPUs and an NVIDIA Tesla C2070 GPU.
• Simulated population PK data using two different PK models with rich/sparse sampling schemes.

Main Results:

• The MCPEMGPU consistently achieved shorter computation times than MCPEMCPU.
• A speedup factor exceeding 48-fold was observed using a single GPU card.
• Parameter estimation and model computation times were significantly reduced with the hybrid approach.

Conclusions:

• The hybrid GPU-CPU MCPEM algorithm significantly accelerates population PK data analysis.
• This novel implementation shows great promise for next-generation population PK/PD modeling software.
• GPU-accelerated computing can substantially improve the efficiency of model-based drug development.