Two Sequential Monte Carlo Samplers for Exact and Approximate Bayesian Inference on Cardiac Models

This tutorial describes how to download and execute the Sequential Monte Carlo code for inference on the linear, polynomial, and O’Hara-Rudy models as described in “Two Sequential Monte Carlo Samplers for Exact and Approximate Bayesian Inference on Cardiac Models”, Accepted by Journal of the Royal Society Interface in 2017.

Installation#

This project requires the Functional Curation add-on to Chaste in order to run, which in turn requires the Chaste source tree to be installed. Instructions for this installation can be found for a variety of operating systems under InstallGuides/.

Extra install commands needed as well as the Ubuntu chaste-dependencies package are (on 14.04 at least):

#!sh
sudo apt-get install python-dev python-scipy python-numpy cython python-tables python-matplotlib python-numexpr python-pip
sudo apt-get install scons
sudo -H pip install dill pathos

Afterwards, obtain the latest version of all the code from the Chaste repositories using:

#!sh
git clone -b develop https://chaste.cs.ox.ac.uk/git/chaste.git Chaste
cd Chaste/projects

## Use your email address as the password for the 'anonymous' account.
svn co https://chaste.cs.ox.ac.uk/svn/chaste/projects/FunctionalCuration --username anonymous --password my.email@domain.com
svn co https://chaste.cs.ox.ac.uk/svn/chaste/projects/DalySMC --username anonymous --password my.email@domain.com

Usage#

Source code for the SMC parameter fitting algorithms is contained in the src folder. Python scripts for performing inference on all model problems can be found in tests.

An annotated CellML file describing the O’Hara-Rudy model problem can be found in the top-level project directory as ohara_rudy_2011.cellml, while functional curation protocol files that describe the simulation and recording thereof are contained within tests/protocols.

A description of important files and their contents follows below.

Files relevant for implementation of the samplers:

• src/python/modeling/fitting/approximatebayes contains the implementation of both Toni and Del Moral SMC algorithms.
• src/python/modeling/fitting/objective/py contains the implementation of the objective functions used for inference

Files relevant for inference on the linear model:

• test/LinearFitting.py defines the linear model and all fitting experiments associated with this model as described in the paper.

Files relevant for inference on the polynomial model:

• test/PolynomialFitting.py defines the polynomial model and all fitting experiments associated with this model as described in the paper.

Files relevant for inference on the O’Hara-Rudy model:

• ohara_rudy_2011.cellml in the top-level directory contains a CellML description of the equations governing the O’Hara-Rudy model.
• test/protocols/oh_aptrace.txt contains a functional curation protocol that describes the recording of a single action potential from said model.
• test/protocols/oh_aptrace_sumstats.txt contains a functional curation protocol that describes the recording of five summary statistics of a single action potential from said model.
• test/data/OHrAPtrace-8-2.dat contains simulated noisy action potential data that is used for exact inference in the study.
• test/data/OHrSumStats-8-2.dat contains sumulated noisy summary statistic data that is used for approximate inference in the study.
• test/OHaraRudyFitting.py defines all fitting experiments associated with this model as described in the paper.

To generate all described posterior estimates for the linear model as described in the paper, go to the top-level Chaste directory and use:

scons projects/DalySMC/test/LinearFitting.py

To generate all described posterior estimates for the polynomial model as described in the paper, use:

scons projects/DalySMC/test/PolynomialFitting.py

To generate all described posterior estimates for the O’Hara-Rudy model as described in the paper, use:

scons projects/DalySMC/test/OHaraRudyFitting.py

To see verbose output on the progress of the ABC algorithm, add the flag no_store_results=1 to the scons commands above. Note however that this will prevent storing a copy of the output on disk.

Section contents#

SubWiki()