MRC Doctoral Training Partnership: Artificial Intelligence Methods for Drug Discovery

Ion channels exist in all human cells and produce tiny electrical currents typically recorded with the Nobel Prize winning “patch-clamp” technique. Loss or dysregulation of ion channels underlies many diseases; including “LQT” associated Sudden Death. Interest from the pharmaceutical industry is great since ion channels are targets for many drugs and the Global market for ion channel modulating medicines exceeds $11.5bn.

Hypothesis
Artificial Intelligence and automated patch clamp can accelerate the discovery of drug modulation of ion channels.

Detailed analyses of ion channel activity is critical in drug development both in screening for novel ion channel drugs and screening against potentially fatal drug side effects. New technology is now delivering automated ion channel recording systems. Manufacturers include Nanion and we have one of these machines termed a high through-put screening (HTS) device. The general problem with these units is that there are no software solutions that can fully analyse all the ion channel data that these machines can produce.

Artificial Intelligence is the key to solve this problem. We propose to prototype models to automatically process large collections of ion channel data using a type of artificial intelligence called “deep learning”. Deep learning offers the potential to solve complex and ambiguous problems. Over the past few years, a plethora of deep learning tools such as convolutional neural networks (CNNs), recurrent neural networks (RNN) including long short-term memory (LSTM) have been developed. These tools, either used alone or in combinations, have shown better performance than human experts in different applications such as computer vision and speech recognition. Further development of deep learning techniques could have a profound impact on ion channel research.

Project Outline
Potentially, activation of large calcium activated potassium ion channels (BK) in blood vessels could be a useful treatment for hypertension, conversely off target block of this channel could lead to dangerous hypertension. Therefore, the student will focus artificial intelligence research on BK channels in vascular smooth muscle. The student will learn tissue culture and patch-clamp of vascular smooth muscle. They will identify BK channels and calculate an IC50‘s for these channels with such modulators, such as iberiotoxin, TEA, NS004 or novel mediators from the FDA drugs databank. They will benchmark the sensitivity and speed of ion channel modulation detection.

The next phase of the project will be to develop the AI tools. We currently have prototypes which can detect clear ion channel activity, and now the student will refine these models to identify ion channel events and measure activity in real cells. Students will also apply visualisation techniques for deep learning to extract detailed information from the hidden layers of the deep learning model.

Finally, the successful candidate student will deploy the AI models to detect ion channel modulation of BK in “cells” with automated multi-headstage patch-clamp Nanion Patchliner apparatus, test channels for sensitivity to drugs using new deep learning models before finally quantifying performance; detection speed and sensitivity against the benchmark set at the start of the project.