How does epigenetics affect immune disease?
I believe that in most case it is unclear how epigenetics exactly affects immune diseases. Most studies to date have associative in nature, meaning that epigenetic differences are correlated with the occurrence, or particular phenotypes, of immune mediated inflammatory disorders (IMIDs). That being said, at the level of histone modifications, there is some evidence that epigenetics may actually modulate the inflammatory phenotype through particular epigenetic readers and erasers. One of the biggest problems in understanding how epigenetics affects the pathogenesis and/or etiology is due to the fact that epigenetics is a cell-type specific feature with most studies interrogating heterogeneous populations: any “epigenetic” difference could be the result of population differences and/or actual differences at the level of the epigenome. This is exactly the question we try to elucidate in work package 5 where we seek to disentangle the DNA methylome from cellular heterogeneity.
Why is this research novel and does it add to existing techniques?
The current research is novel in that the use of DNA methylation as biomarker for predicting response to therapy has not been implemented outside the field of oncology. Beyond the DNA methylation analyses on peripheral blood, the characterization of DNA methylation on single cells is not an experiment that is done routinely. If successful, our data would not only show that the implementation is possible in peripheral blood, but also provide a resource for other researchers to interrogate the DNA methylome at single cell level.
What is an AI algorithm?
An AI algorithm can be a very broad term. The way we are implementing our algorithm is essentially a set of rules that define, based on the DNA methylome, whether the provided DNA methylation data from a particular sample resembles more a responder or non-responder. The AI algorithm is in our case a (set of) decision tree(s), which essentially boils down the same principle used by daycares in the Netherlands during COVID-19 when deciding whether a could go to the daycare. Importantly, the actual AI algorithm does not need to be the most complicated (although it can certainly be!) but rather the process by which one arrives at such an algorithm. This process is, just like the rest of science, an iterative process by which the most optimal algorithm is decided upon through multiple rounds of training and validation.