Bayesian inference for fitting cardiac models to experiments: estimating parameter distributions using Hamiltonian...

Area of Science:

• Computational biology
• Biophysics
• Cardiovascular research

Background:

• Patient-specific cardiac action potential models are crucial for predictive tools.
• Traditional optimization methods struggle with noisy data, yielding single parameter fits.
• Existing Bayesian methods like Markov chain Monte Carlo are computationally inefficient.

Purpose of the Study:

• To evaluate two computationally efficient Bayesian approaches, Hamiltonian Monte Carlo (HMC) and approximate Bayesian computation sequential Monte Carlo (ABC-SMC).
• To assess the effectiveness of HMC and ABC-SMC in customizing cardiac action potential models using synthetic and experimental data.

Main Methods:

• Utilized Hamiltonian Monte Carlo (HMC) algorithm for parameter estimation.
• Employed approximate Bayesian computation sequential Monte Carlo (ABC-SMC) algorithm for parameter estimation.
• Applied both methods to two cardiac action potential models with synthetic and zebrafish experimental data.

Main Results:

• Both HMC and ABC-SMC successfully identified distributions of model parameters for cardiac action potential models.
• HMC generally produced narrower marginal distributions compared to ABC-SMC.
• ABC-SMC demonstrated less sensitivity to algorithmic settings, including prior distributions.

Conclusions:

• HMC and ABC-SMC are computationally efficient and effective Bayesian methods for customizing cardiac action potential models.
• These methods enable the development of more accurate patient-specific models and virtual patient cohorts.
• The choice between HMC and ABC-SMC may depend on specific data characteristics and desired precision.