Back to blog

How can we… use AI to advance metabolic medicine?

Christopher Bannon, Registrar Clinical Research Associate in Metabolic Medicine, University of Cambridge

19 February 2024

Categories:

Accelerate spark data science residency

Categories:

Machine learning

We are currently battling an obesity pandemic, with worldwide rates almost tripling since 1975 . There is understandably a large and urgent interest in better understanding the underlying mechanisms that contribute to obesity, including how the gut hormones may be involved. Analogues of the gut hormone GLP-1 such as Ozempic and Wegovy are now used to treat both diabetes and obesity.

My research looks at how gut hormones levels and other markers of metabolism change in healthy individuals compared to patients with obesity, and also compared to patients with bowel conditions which impact 6.5 million people in the UK.

My research

I am a medical doctor more than halfway through my PhD in gut hormone physiology in the Gribble-Reimann group at the Institute of Metabolic Science, University of Cambridge. I study gut hormones involved in appetite, satiety and motility. I study levels of certain gut hormones in the fed and fasted states in healthy people and those with different bowel conditions, recruiting participants and using new techniques in the lab to measure levels of different hormones and other metabolic markers. I regularly collaborate with the clinical endocrinology and gastroenterology teams at Addenbrooke’s Hospital. I aim to characterise these levels in healthy volunteers and look at how they vary in different disease states.

Most people in my field of research look at gut hormones and how they are connected to diabetes or obesity because they are appetite and metabolism, but I’m predominantly interested in what regulates the release of hormones that either slow down or speed up processes in the gut.

There are around 20 hormones produced in the gut. The most studied gut hormone to date is glucagon like peptide-1 (GLP-1) which is involved in controlling blood glucose and appetite, but also slows down how food move through the gut. GLP-1 analogues are now used for treatment of obesity, but has also been additionally recognised recently a treatment for patients with bile acid diarrhoea (BAD). I not only study GLP-1, but other much less studied hormones including motilin and insulin-like peptide 5 and how they are involved in gut motility.

One condition I have particularly focused on is bile acid diarrhoea (BAD) that affects around one million people in the UK. Bile is known to stimulate lots of different hormones from the gut, so we have chosen to focus on this condition to try and understand these hormones in more detail, and how their hormone levels compare to individuals with other bowel conditions, and healthy volunteers.

Our group in collaboration with Richard Kay uses liquid chromatography mass spectroscopy for greater specificity and the opportunity to test for novel biomarkers, as well as data from other studies, which together could potentially aid in the diagnosis and treatment of different gastrointestinal and metabolic diseases including BAD.

Making data more digestible

As part of my research, I wanted to integrate different data sets to explore measured gut hormones, demographics features and metabolic markers across different conditions, but it proved difficult to merge and analyse without a data science background.

So, to develop the skills I needed, I took the Accelerate Programme’s 5-week Introducing Data Science for Research course and learned the basics of Python, which I enjoyed. It allowed me to learn and master the basics of data preprocessing and merging using the pandas and numpy packages, alongside being to produce much higher quality data visualisations using the seaborn package. This has elevated my work by merging databases effectively and doing quick high quality visualisations to present data in more aesthetic ways.

I then took the Accelerate Programme’s year-long Machine Learning Academy. I am using the skills I learned to take my data further. I have so far used unsupervised and supervised algorithms to explore predicted gut hormone levels in obesity, whilst using ensemble techniques to explore proteomic datasets to find novel biomarkers in different bowel conditions. I hope to be able to predict whether someone’s gut hormone or metabolic marker levels are that of a healthy person, or whether they have a metabolic or bowel condition.

Since completing the course, I have attended a Machine Learning Engineering Clinic with Accelerate’s Machine Learning Engineer Ryan Daniels to sense-check my data techniques with an expert, as the majority of my group are clinical or biological scientists with less experience in computer science. Ryan helped me further develop my data science skills, and has provided expertise in feature selection processes to help find import biomarkers in my datasets.. We are collaborating on a forthcoming project involving big databases incorporating different conditions and hundreds of different protein fragment readouts. With the techniques Ryan has shown me, I should be able to see which fragments cluster in one set versus another.

The future

I will be presenting some of my findings at the Digestive Diseases Week in USA later this year. I feel I have found my niche – looking at gastroenterology from a hormone point of view – and plan to continue looking at gut hormone levels and try to unravel the sort of dysfunction that occurs with different conditions. I would also like to continue following a big data approach as well as seeing individuals in the clinic. It’s an exciting area. Currently, people look to diet or their microbiome to relieve symptoms of IBS and other issues, but I think there is a piece of the puzzle missing: gut hormone levels.

I hope to carry on collaborating to help find this missing piece and use big data, machine learning and AI to find new biomarkers to diagnose conditions or suggest if one treatment would work better than another. I hope to take what I’ve learned on the Accelerate courses to address those questions.

Find out more about the data science and machine learning courses offered by the Accelerate Programme on our resources page or get in touch at accelerate-science@cst.cam.ac.uk.

Accelerate’s Machine Learning Engineering Clinic is available to all researchers at the University of Cambridge to support you in implementing machine learning at all stages of the research pipeline. Find out more and contact us.