Machine Learning in researching autoimmune disease

The Immunology research team at University College London (UCL) hosted a public information event this week, hosted entirely online due to the challenges imposed by COVID-19. Having had the privilege of attending, we have provided a very brief summary of the fascinating event below. It was presented live through Zoom, with polls, feedback and Q&A via sli.do (allowing participants to interact with the presenters).

Hosted by George Robinson of UCL Immunology, and introduced by Prof Liz Jury, we heard what machine learning is and how it has been used in current research projects.

What is machine learning?

Despite Machine Learning being full of technical buzz words, Leda Coelewij from UCL explained what it is using every day terms and examples. In short, Machine Learning is about using computers to look at lots of data at once (sometimes known as 'big data'). The computers use mathematical tools to build models. The idea is that they develop a model that allows researchers to split things into categories. For example, using measurements of an iris flower to identify which particular variety it is - each variety is very slightly different, and the computer can build a model to identify them.

There are two types of machine learning. The first is 'supervised', where you teach the computer what something is and give it all the data you hold. The computer builds a model, and you can then test it using new examples and you can then assess the accuracy of the model - the more accurate it is, the better the model is at correctly predicting the correct group. The second type, which is used less often, is 'unsupervised' where you don't tell the computer the groups at all and leave it to work out its own groups.

There are different machine learning techniques, including 'K nearest neighbours' (grouping based on information you have, so any new cases are then put into the group they are closest to), decision tree (where the computer decides how to split categories to create final groups), random forest (lots of decision trees used to create a final model), and logistic regression (looking at how different pieces of information relate to the groups, and giving each a 'weight' or level of importance).

Multiple Sclerosis is an autoimmune disease and, like JIA, many patients are given disease modifying drugs. In some cases, patients develop antibodies to the drugs, which can lead to them not working. A machine learning model using biomarkers (information from blood samples) was developed by UCL Immunology to predict which patients will likely experience this, and the model was correct around 84% of the time. It also helped identify which biomarkers are most useful in predicting disease response.

Machine learning in JSLE

Junjie Peng (UCL) presented some of his work in Juvenile Lupus (JSLE). In that study, blood samples from JSLE patients and healthy controls were analysed. A total of 28 different immune cell types were analysed, and machine learning was used to identify which are most important in JSLE. They found patients with JSLE fell into 4 different groups, and 8 of the immune markers were important in identifying them. This has been used to predict long-term outcomes in different groups, and the hope is it will lead to better targeted treatments. Similar work has been undertaken within the CLUSTER study into JIA.

Machine learning in disease research is an exciting and developing area. The research will help in identifying disease subtypes or patients most likely to respond to certain medications. Further work to help target medications more specifically will take time, but will ultimately help patients with a range of autoimmune diseases.

Thank you

UCL Immunology are planning further seminars for patients and parents, so keep an eye out. A recording of the session this week will be uploaded to their YouTube channel. A huge thank you to George, Leda and Junjie for taking the time to explain machine learning and how it is in use in medical research, as well as to Prof Liz Jury for introducing the event and the work of her team.