High-Performance Computing Frameworks for Bio-Simulation in Regenerative Sciences

Abstract

Bio-simulation in regenerative sciences -- encompassing ODE pathway kinetics, agent-based tissue simulations, finite element mechanics, reaction-diffusion PDE systems, and multi-scale coupled models -- generates some of the most computationally demanding workloads in computational biology, with individual CITRA agent-based simulations requiring 8 hours on 32-core CPUs and TEHMC coupled PDE-ABM models requiring 8.4 hours per design evaluation. The transition from CPU-only to GPU-accelerated, and from single-node to distributed high-performance computing (HPC) architectures is essential for making these simulations tractable at the throughput required by RPSO protocol optimisation (200-500 evaluations per indication) and PRDT digital twin personalisation (50 TRS-sim evaluations per patient). This paper proposes the Regenerative Sciences High-Performance Computing (RSHPC) framework, a systematic methodology for GPU acceleration and distributed HPC implementation of regenerative biology simulation workloads, comprising four technical components: a GPU-accelerated ODE solver suite (GAOS) implementing RPSSB pathway models on GPU; a parallel agent-based simulation engine (PABSE) implementing GPU-parallelised CITRA-compatible ABM; a distributed PDE solver framework (DPDF) scaling TEHMC NTRPS and SMM FEniCSx solvers to multi-node HPC clusters; and a heterogeneous workload orchestrator (HWO) managing mixed CPU/GPU/HPC workloads for bio-simulation pipelines. RSHPC is benchmarked on RPSSB, CITRA, and TEHMC workloads across CPU, GPU, and multi-node HPC configurations. GAOS achieves 148x GPU speedup over CPU-only ODE integration for RPSSB pathway models. PABSE achieves 84x GPU speedup for CITRA ABM (100,000-agent simulations). DPDF achieves 28x speedup for TEHMC NTRPS on 32-node MPI cluster vs. single-node FEniCSx. HWO reduces total RPSO optimisation wall time from 280 hours (CPU-only sequential) to 2.8 hours (GPU+HPC parallel). The study contributes the RSHPC specification, GPU/HPC implementation guidelines, and empirical benchmarks enabling the regenerative biology simulation ecosystem to operate at practical throughput for protocol optimisation and digital twin personalisation.