GpABC

GpABC implements approximate Bayesian computation (ABC) in Julia for likelihood-free parameter inference and model selection of systems biology models characterized by stochastic and nonlinear dynamics.

Key Features:

• Likelihood-Free Inference: Enables parameter estimation without explicit likelihood calculations for models with intractable likelihood functions.
• Model Selection: Supports comparing and selecting among competing biological models within the ABC framework based on fit to observed data.
• ABC-SMC and Rejection ABC: Implements standard rejection ABC and Sequential Monte Carlo (ABC-SMC) methods for simulation-based parameter inference.
• Gaussian Process Emulation: Uses Gaussian process emulation to approximate simulation outputs and reduce the computational cost of ABC.

Scientific Applications:

• Parameter inference in systems biology: Inferring parameters for stochastic and nonlinear biological models where likelihoods are intractable.
• Model selection for competing hypotheses: Comparing mechanistic biological models using ABC-based model selection.
• Analysis of computationally expensive simulations: Enabling ABC on slow stochastic simulators by reducing simulation load via Gaussian process emulation.

Methodology:

Standard rejection ABC and ABC-SMC (Sequential Monte Carlo) compare simulated data to observed data and accept parameters that produce sufficiently close simulations. ABC with Gaussian process emulation approximates model simulation outputs with a Gaussian process to accelerate the ABC procedure.