Identifying nutrient and pesticide modulation of ion channels with Artificial Intelligence (AI)

Ion channels govern many physiological processes in humans and animals, and subtle changes in their behaviours can have dramatic biological effects - for example, common pyrethroid insecticides frequently enter the food chain and cause toxicity via interactions with ion channels (especially GABA receptors). In extreme cases, the consequences of these interactions can be fatal, such as in LQT associated sudden death. Even glucose, artificial sweeteners and fatty acids can have direct effects on ion channels in the heart, predisposing individuals to arrhythmia. Understanding how these interactions work is crucial for treatment of ion-channel related conditions (known as channelopathies), as well as developing safer pesticides, drugs and food for the future.

Typically, the impact of different drugs on ion channels is measured via the Nobel prize winning patch-clamp technique - ion-channels produce tiny electrical currents (~1 picoamp) across the plasma membrane, and changes in current correspond to an opening or closing of either a group of, or a single ion channel. While strides have been made increasing the throughput of measuring these effects, the analysis currently needs extensive experimenter supervision, raising issues around subjectivity. It is also thought that for single channel analysis, channel behaviour is best described using an underlying Hidden Markov Model (HMM), rather than a simple open/closing model. This can involve a single channel having many "closed" and "open" states, each pertaining to a different physiological configuration within the cell membrane. While human-supervised software exists for idealisation and analysis of ion-channels and their related HMMs, there is no known unbiased, automated package for Markovian analysis of ion-channels - this project aims to develop novel mathematical techniques to create such models using emerging technology.

Deep Learning (DL) - a branch of artificial intelligence - has proven to give better performance than human experts in many different applications, such as computer vision, language interpretation and anomaly detection. Improvements in computational power, as well as improvements in convolutional neural networks (CNNs), and recurrent neural networks (RNNs) - specifically long short-term memory (LSTM) models have led to these types of models making a profound impact on industry, in particular in pesticide and nutrition R&D.