Modelling drug binding to biological ion channels

Frankie Patten-Elliott

Modelling drug binding to biological ion channels

Publication Date

July 30, 2026

Creator

Frankie Patten-Elliott

University of Nottingham

Abstract

Ion channels are proteins that play a crucial role in many biological functions. In a healthy heart, ion channels in cardiac muscle cells ensure the heart pumps in a regular, coordinated manner. Pharmaceutical drug compounds are prone to binding to cardiac ion channels, disrupting healthy cardiac function and sometimes leading to the onset of cardiac arrhythmias. During drug development, significant time and money are spent on cardiac safety testing to avoid these potentially fatal side-effects. Much consideration is, therefore, given to improving cardiac safety testing methods to reduce uncertainty in risk predictions, while minimising time and cost. Mathematical models can provide key insights into the underlying mechanisms that define complex biological systems. In the field of cardiac electrophysiology, models of ion channel gating and drug binding mechanisms can be effective tools for predicting drug-induced proarrhythmic risk. In this thesis, we consider methods to improve models of drug binding mechanisms, with a specific focus on binding in the human Ether-à-go-go-Related Gene (hERG) channel. Recent technological advances have enabled the collection of high-frequency electrophysiology data that can be used to calibrate models of drug-channel binding. However, careful consideration must be given to ensure the collected data are sufficiently information-rich to discriminate between different proposed models of binding mechanisms. We present an approach that produces optimal experimental designs to aid model discrimination, thereby uncovering drug-specific mechanistic insights and assisting in cardiac risk assessment. By developing improved model-fitting methods, we also overcome some of the limitations introduced by experimental artefacts in the collected data, which, if left unaccounted for, can bias model fits and predictions. We then investigate Machine Learning (ML) methods to learn model equations directly from data. These methods allow greater modelling flexibility by significantly extending the scope of potential model structures. We demonstrate that the effectiveness of the considered ML methods is dependent on sufficiently informative data and ample prior knowledge of the missing dynamics, suggesting that combining ML methods with experimental design techniques could prove more successful.

Item Type

ethesis

Thesis Type

PhD

Supervisors

Mirams, Gary R.

University of Nottingham

Preston, Simon P.

University of Nottingham

Wilkinson, Richard