This poster was presented at "Fickle Heart: The intersection of UQ, AI and Digital Twins" on 3rd June 2024.

Abstract

Development of mathematical models for ion channels typically requires estimation of parameters from experimental data. Accurate identification of the optimal parameters is necessary for model results to be generalisable. However, the choice of optimisation approach can lead to sub-optimal parameters, excessive use of computational resources, or both.
Many optimisation approaches have been developed for and used on ion channel models, but a lack of comparisons against standard problems make the choice of a suitable approach unclear. Optimisation benchmarks have the potential to give insight into such a choice, but no existing benchmark is specific to ion channels.
We have developed an open source benchmarking tool, using ion channel optimisation problems from the literature, allowing standardised comparisons between existing and future optimisation approaches. Using this tool, we benchmarked the performance of over thirty unique approaches that have been previously applied to ion channel models. Finally, we used the results of this benchmark to derive standard recommendations for ensuring the best parameters are achieved at the lowest computational cost.