Unravelling the complexities of the immune system with AI

immune system

AI is helping to broaden our understanding of the immune system, including how it responds to and stores information on external pathogens and diseases. Johannes Textor from Radboud University in the Netherlands tells us more.

Disruptive technologies already significantly impact the way we monitor, diagnose, and manage infections and diseases. Of course, a key component in determining the trajectory of any disease or infection within the body is our immune system. Made up of a complex network of organs, tissues and cells, it processes and stores information on invading pathogens to better protect the body from future invaders.

Advances in technology are enabling scientists to harness the power of Artificial Intelligence (AI) and computational modelling to better understand the mechanisms of this vast and somewhat enigmatic system in the hope that it could inform the development of more effective diagnostics and therapeutics.

To understand how scientists are attempting to replicate certain aspects of the immune system artificially, Lorna Rothery spoke to Johannes Textor who leads the computational immunology group at Radboud University in the Netherlands. His research focuses predominantly on the design and application of simulation models and Machine Learning to examine information processing in the adaptive immune system.

How did the field of computational immunology come to be established and what does it entail?

The field of computational immunology has a long history that really took off in the early 1990s as part of biosecurity research taking place in Los Alamos in the US. Biophysicist Alan Perelson, now a senior fellow of the Los Alamos National Laboratory, was working on quantitative and computational models of the immune system, which at that time was really novel. Researchers collaborated with computer scientists to broaden their understanding of the immune system while also using that research as a template to build safer computer systems that were protected from viruses. Papers were published on the use of computational models to understand the immune system and that same research would also inform the creation of computer algorithms to protect the machines from viruses.

People were initially enthusiastic about the idea of artificial immune systems and more and more groups around the world joined this research effort; there was even a dedicated conference set up which had not been done before. However, as people started to realise how complex the immune system is, and struggled to progress their research beyond the prototype stage, momentum in the field faded.

Since then, the field of computational immunology, which incorporates the statistical analysis of immunological data, has become more mainstream, but using the immune system as a template for computer algorithms is still very niche.

How do you approach the design of artificial immune systems? How can they help us to predict how the body will respond to a particular infection or vaccination, for example?

The immune system can react to the same stimuli in different ways and learn from past experiences, so if you contract an infection for the second time, your response to that infection will be different. How systems are designed depends on the research question you are trying to answer. As well as being a complex system consisting of interacting chemicals, the immune system also contains various subsystems which add to the complexity of the design. Some people opt for a broader and more holistic approach when building an artificial immune system, but this is an ambitious undertaking given its complexity; I do not think we yet know enough about the system and its processes to produce a realistic representation. Others have produced very simple – but equally useful – models of the immune system that consist of a single mathematical equation. Alan Perelson, for example, wrote a famous paper where he used simple mathematical models to predict how people with HIV would react to antiretroviral treatments and if they would need to take those treatments for the duration of their life.

My team and I are focusing on T cells and building models of the T cells’ arm of the adaptive immune system, which represent the size and complexity of the T-cell receptor (TCR) repertoires. The innate arm of the immune system consists of neutrophils, macrophages and other cells which, while playing an important role in protecting the body from invading pathogens, do not exhibit the same advanced learning mechanisms as T and B cells. We aim to eventually predict T-cell responses, but these responses also involve molecular interactions between T cells and epitopes, which are hard to predict. At present, we cannot tell from looking at a sequence of a T cell receptor which epitope it is going to react to. Artificial Intelligence, combined with more data, would help to solve this problem.

In principle, I hope that at some point, we might know enough that we can make models that can predict immune responses, in a similar way to how a model predicts whether it is going to rain, or the trajectory of the climate based on what information is available, and by simply simulating the process. Everyone has unique T-cell repertoires, so there is quite a lot of information that you could potentially use to better understand why certain people do or do not react to certain treatments, for example. In cancer, we see a lot of heterogeneous responses to the same treatment, and we want to understand why; part of it is due to the complexity of the immune system and the differences between people.

How does the immune system differ from other information-processing systems like the brain and central nervous system? Equally, how do these systems interact with each other?

In the brain, we have a network of similar types of cells with varying complexities and connections. This neural network is what encodes the complexity of the system, and that works like a circuit, computing various functions. We can now make powerful Machine Learning systems based on that paradigm, but the immune system is very different. It consists of single cells that are more distributed than neurons and do not make static connections in the same way. The complexity of immune cells on the genetic level is much greater than that of neurons, though the immune system as a whole is perhaps less complex than the central nervous system. There has been a lot of research on the interactions between the central nervous system and the immune system – there is even a whole dedicated subfield called neuroimmunology which operates at the intersection.

Research shows that you can offset immune ageing in part by having cells that are more experienced and therefore better able to fight the most common pathogens ©shutterstock/piccreative

Immune cells have sophisticated genetic machinery that can only be found within these cells, and every cell gets a new receptor made specifically for it. For example, there is a high chance a single T cell from my body has one receptor that is unique to that cell, that no other cell in my body carries. That is really where the enormity and complexity lie. The way that these cells interact and learn is very different from how the brain learns. People have been studying this in the context of learning classifier systems and making abstract models of agents that are both simple, and very different. They learn collectively by reinforcing the agents that make the correct decisions.

I have worked with neuroscientists to study sleep and the impact of sleep on immune systems. We can see that sleep helps your brain to learn, but it also helps the immune system to learn. People know that sleeping well is quite important for your overall health, and it has been demonstrated in the lab. There was a randomised control trial where participants received a vaccination and half of them were kept awake while the other half went to sleep as normal, you can still detect the difference in immune system response a year later.

There is crosstalk between the systems but equally, there may be other processes happening during sleep that have an impact, such as changes in hormone levels; the mechanisms are not fully clear. It is still an exciting area of research but is also fairly niche because you have to understand both of these systems; it is quite a feat.

What about the roles of the immune system and ageing in the development of chronic diseases like cancer?

Certain things happen to the immune system as we age, over the life course your immune system diversity shrinks because you are not producing as many new cells. The thymus gland which produces white blood cells starts to shrink, from as young as 14-18 years old. This process is called immunosenescence. Research shows that you can offset immune ageing in part by having cells that are more experienced and therefore better able to fight the most common pathogens. Part of my research involves supporting studies focused on cancer treatments, by designing them in a better and more efficient way that makes it possible to find effective treatments.

Due to the wealth of data that we have, medical research can become data science, in effect. A lot of PhD students, even in clinical departments, are collecting data and analysing this, whereas 20 years ago their primary focus might have been merely gathering that data. Now you can collate a lot of data relatively easily, but few people can analyse it well. There is a huge need for an intersection of PhD students who have both lab skills and computational skills, or interdisciplinary teams working together.

What do you think are the biggest issues faced when developing disruptive technologies for the healthcare sector?

Historically, when it comes to the development of new treatments, the return on investment for R&D is shrinking. This is partly because the low-hanging fruits in the context of new medicines have been picked, and developing treatments for more complex, or rarer diseases is more difficult. Another challenge is the increasing amount of complexity and information that we have to deal with in the clinical space when collecting measurements, such as for single-cell RNA sequencing, or advanced imaging technologies. There are lots of people who are analysing very complicated data but are not really qualified to do so; they are not trained data scientists so might create findings that turn out to be a fluke or just the result of how a certain analysis was carried out. Underestimating this aspect, in my view, is one of the biggest challenges.

The culture of academia, whereby there is pressure to publish high-impact papers quickly, as we saw with the burst of papers trying to apply Machine Learning to predict treatment responses, can often dilute the quality and validity of the science we are doing. International projects and multinational funding schemes are a good opportunity to bring people from different fields and with varied expertise together, it is not uncommon now for people to have expertise in both experimental and computational science.

Is AI and data science likely to play a larger role in healthcare in the future?

Lots of people have unrealistic expectations of what it actually means to use AI in healthcare settings, for instance, that the doctor will be replaced by computer programming. When we talk about using AI in healthcare, we are likely not going to replace people with machines but make new software that will support healthcare professionals and hopefully alleviate their administrative burden.

While new technologies are progressing quickly, there is not always a clear pathway for how they will be integrated. AI researcher Geoffrey Hinton predicted seven years ago that we would no longer need radiologists because technology would surpass their abilities and replace them, but clearly, that has not been the case. Yes, we can make computer programmes that can identify tumours and support pathologists to do their job faster and more effectively, but still, you need humans in the loop. We need to recognise that the level of sophisticated technology already being used in the clinic is extremely high.

Johannes Textor
Associate Professor (Universitair Hoofddocent)
Institute for Computing and Information Sciences Radboud University & Radboud Institute for Molecular Life Sciences, Tumor Immunology department, Radboud University Medical Center

This article is from issue 25 of Health Europa Quarterly. Click here to get your free subscription today.

Subscribe to our newsletter


Please enter your comment!
Please enter your name here