The immune system stores enormous amounts of health and disease-relevant information. Researchers are attempting to decipher this information with the help of artificial intelligence, with the aim to develop novel diagnostic and therapeutic methods.
There is still a lot we do not know about how the immune system works. We divide the immune system into different parts. One of them is the adaptive immune system. It develops over the course of our lives. This part of our immune system stores information about all the diseases and infections that a person has or has had. This information is stored as complex patterns in pathogen-recognizing structures, which are also called immune receptors. These are located on the surface of adaptive immune cells.
We can think of these patterns as the memory of the immune system. The patterns are instruction manuals that tell the immune system how to attack various infections and diseases. However, nobody knows what these patterns look like. The potential for what they can tell us about the inner workings of the immune system, as well as for the development of disease diagnostics and therapeutics, is enormous. So, how can we find the patterns?
This is where machine learning comes into play. Machine learning is a form of artificial intelligence. With the help of machine learning, we can let a computer discover the previously hidden patterns for us.
Associate Professor Victor Greiff from the Institute of Clinical Medicine and Professor Geir Kjetil Sandve from the Institute of Informatics are working on this. Together with PhD students Milena Pavlović and Lonneke Scheffer, they have developed the software ImmuneML for that purpose.
– We can use machine learning to find the disease-relevant immune patterns, without knowing what they look like or what characterizes them. That is what is so unique and exciting about machine learning, Greiff says.
Immune patterns can tell us if a person is healthy or diseased
The patterns that the immune system has stored can tell us about a person’s previous and/or current diseases or infections. Greiff, Sandve and colleagues are now trying to figure out which patterns belong to which diseases and infections. If they manage to find out, it can provide new and important knowledge about adaptive immunology.
Importantly, it can also make it easier to diagnose various diseases as well as develop novel therapeutics.
– If we manage to find the patterns, then we may be able to diagnose a number of diseases with a single blood sample, Greiff explains, and adds:
– The patterns can tell us whether the person is healthy or ill, and which disease or diseases the patient may have.
The goal is to find the patterns of thousands of diseases
Greiff and Sandve collaborate with researchers and clinicians in various medical disciplines. They are currently trying to find the patterns of coeliac disease and type I diabetes. For this, they collaborate with researchers at UiO and at University of Florida.
– The real value will unfold when we have learned the pattern for a large number of diseases. In principle, you can then diagnose thousands of diseases from just one single blood sample. That is the goal, Sandve points out.
Can we use machine learning to find the pattern of COVID-19?
If you want to find something, without knowing what you are looking for, you have a difficult task ahead of you. Let us say that we want to find the pattern for COVID-19. How do we know which pattern it is among the millions of patterns that the immune system has stored?
It is a bit like looking for one particular snowflake among millions of other snowflakes. Without knowing what the snowflake that we are looking for looks like.
However, with machine learning, it becomes a completely different matter. Then we can first let the computer find the patterns of a person with a confirmed COVID-19-infection. In this way, the computer “learns” what the pattern for COVID-19 looks like.
– This is how machine learning works. We must first teach the computer what is what. We can do this by letting it find the patterns of a person we know is healthy, and of a person we know to have a specific disease, Greiff explains.
ImmuneML will make machine learning available to more researchers
Initially, Sandve and Greiff’s plan was to develop a software that would make it easier to apply machine learning in their own research. However, they soon realised that there was a need for a common software in the field.
– It felt a bit like we had to dig out the construction site of the new building by hand every single time, before we could start with the construction itself, i.e. the analyses, Sandve says.
Greiff explains that previous studies that have applied machine learning have been of an explorative nature. They have also been poorly standardised and therefore difficult to replicate.
– PhD student Milena Pavlović was tasked with developing software underpinnings that would allow her future studies to be carried out more efficiently and more reproducible than was typical in the field. When Sandve and Greiff saw how well the platform came together they thought, why not share it with others? A second PhD student, Lonneke Scheffer, also joined in, and today, more than a year later, the platform is now available for the whole field to use.
– ImmuneML makes it possible to work more uniformly. We can now compare different studies and assess the value of different approaches.
The aim is that researchers with different levels of experience and expertise in bioinformatics and machine learning can use ImmuneML. The software consists of three different variants with different levels of complexity. It also includes a user manual explaining how to use the software in detail.
– We also have our own YouTube channel with explanatory videos. We hope that many researchers who are interested in this will start using ImmuneML, Sandve concludes.
Associate Professor Victor Greiff and Professor Geir Kjetil Sandve have collaborated and worked on the development of ImmuneML since 2018. The researchers at their labs are working interdisciplinary and are now collaborating on most of their research projects. USIT at UiO and Elixir Norway have contributed to the development of the ImmuneML software.