ion-channel-models

Ion channel model tutorial

The ion channel that is used throughout this tutorial is the ion channel which carries the current IKr.

Models

We are interested in selecting the model that 'best describe' the ground truth (model) from our two candidate models.

• Model A (candidate model): Beattie et al. 2018 model.
• Model B (candidate model): Oehmen et al. 2002 model.
• Model C (ground truth model): Di Veroli et al. 2013 model (temperature at 295K; parameters from fitting to Beattie et al. 2018 cell #5 data).

Use of protocol

In this tutorial, we split our protocols into calibration and validation uses.

• Protocol sinewave: calibration.
• Protocol staircase: validation.
• Protocol ap: validation.

Arguments

• [which_model] can be one of A, B.
• [which_predict] can be one of sinewave, staircase, AP.

Run the tutorial

Before calibration, run generate-data.py to generate synthetic data with i.i.d. Gaussian noise (create data).

1. Calibration with i.i.d. noise assumption

1. Run fit.py with argument [which_model] to calibrate the specified model. Alternatively run fit-all.sh.
2. Run predict.py with arguments [which_model] and [which_predict] to predict the specified (protocol) data (in ./data) with the specified model and calibrated model parameters (in ./out). Alternatively run predict-all.sh.
3. Run compare.py with argument [which_predict] to compare the predictions of the specified (protocol) data (in ./data) from the candidate models with the calibrated model parameters (in ./out). Alternatively run compare-all.sh.
4. Run plot-residual-noise.py to plot the autocorrelation of the calibrated models.
5. Run mcmc.py with argument [which_model] to run MCMC for the specified model. Alternatively run mcmc-all.sh.
6. Run plot-iid-sigma.py to plot the marginal posteriors of the i.i.d. noise parameter .

2. Calibration with discrepancy model: GP(t)

1. Run fit-gp.py with argument [which_model] to calibrate the specified model.
2. Run mcmc-gp.py with argument [which_model] to run MCMC for the specified model.

3. Calibration with discrepancy model: GP(O,V)

1. Run fit-gp-ov.py with argument [which_model] to calibrate the specified model.
2. Run mcmc-gp-ov.py with argument [which_model] to run MCMC for the specified model.

4. Calibration with discrepancy model: ARMA(p,q)

1. Run mcmc-arma.py with arguments [which_model], [arma_p], and [arma_q] to run MCMC for the specified model, where [arma_p] and [arma_q] are integers specifying the order of the AR and MA models, respectively. We use ARMA(2,2) model throughout the paper.

5. Run posterior predictive

1. Run posterior.py with arguments [which_model] and [which_predict] to generate posterior predictive with i.i.d. noise assumption. Alternatively run posterior.sh.
2. Run posterior-gp.py with arguments [which_model] and [which_predict] to generate posterior predictive with GP(t) discrepancy model. Alternatively run posterior-gp.sh.
3. Run posterior-gp-ov.py with arguments [which_model] and [which_predict], together with a -ov flag, to generate posterior predictive with GP(O,V) discrepancy model. Alternatively run posterior-gp-ov.sh.
4. Run posterior-arma.py with arguments [which_model], [arma_p], and [arma_q] to generate posterior predictive with ARMA(p,q) discrepancy model. Alternatively run posterior-arma.sh.

Alternatively, run posterior-all.sh to generate all the posterior predictive for all models in one go.

6. Plot final figures (comparison)

1. Run compare-posteriors.py with argument [which_model] to plot the marginal posterior distributions for the specified model, for all discrepancy models. Alternatively run compare-posteriors.sh.
2. Run compare-pp.py with arguments [which_model] and [which_predict] to plot the posterior predictive for the specified model and protocol/data, for all discrepancy models. Alternatively run compare-pp.sh.
3. Run compare-evidence.py with argument [which_model] to generate a table (in PDF/PNG format) showing the marginal likelihood of the specified model, for all discrepancy models and protocols/data.
4. Run compare-error-mean.py with argument [which_model] to generate a table (in PDF/PNG format) showing the error mean of the specified model, for all discrepancy models and protocols/data.
5. Run compare-mean-error.py with argument [which_model] to generate a table (in PDF/PNG format) showing the mean error of the specified model, for all discrepancy models and protocols/data.

Alternatively, run compare-tables.sh to generate all the tables (Steps 3-5) in one go.

Output

• out: Fitting output etc.
• fig: Output generated figures.
• data: Data generated from the ground truth model.

Utilities

• method: Contains all the useful methods/functions for this tutorial.
• mmt-model-files: Ion channel models in Myokit mmt format.
• protocol-time-series: Voltage clamp protocols in csv, time-series format. Each file has two columns, the first one is time (in [seconds]) and the second column is voltage (in [milliVolts]).

Tests

• test-models.py: Simple test for default model setting and forward model simulations.

Others

• fit-gp-v.py, mcmc-gp-v.py: Run calibration with GP(V) discrepancy model, which is not shown in the paper.
• fit-gp-tv.py, mcmc-gp-tv.py, posterior-gp-tv.py: Run calibration and posterior predictive with GP(t,V) discrepancy model, which is not shown in the paper.
• mcmc-arma-invertible.py: Run calibration with ARMA(p,q) discrepancy model and the invertibility condition, which is not shown in the paper.