To the untrained eye, a blood smear under the microscope reveals two types of immune cells: red blood cells and white blood cells. But immunologists know that white blood cells can be further categorized into a staggering array of subpopulations, distinguished by the genes, or markers, they express.
Over the years, scientists have found specific sets of markers, that identify different immune cell populations. But as the library of markers grows, so does the complexity of correctly assigning markers to a particular immune cell subtype, especially when those subtypes may overlap in function and location in the body.
Florent Ginhoux, a Senior Principal Investigator at A*STAR’s Singapore Immunology Network (SIgN), views this complexity as a classification problem that machine learning is well suited to solve. In a study published in the journal Immunity, his team devised a method called InfinityFlow which leverages machine learning to characterize immune cell populations based on 332 markers. Collaborators from Greece, the Netherlands and the US were involved in the study.
“The innovativeness of our approach lies in integrating the data on all 332 markers and analyzing them using a machine-learning algorithm,” Ginhoux explained. “This allowed us to predict which sets of markers best defined which immune subtypes.”
The researchers used their technique to clarify the identities, functions and lineages of two types of immune cells: conventional dendritic cells and classical monocytes. Previously, a marker called CD14 was thought to be a positive and definitive identifier of classical monocytes. However, the analysis by Ginhoux’s team revealed that a subset of cells expressing CD14 were not monocytes, but belonged to the dendritic cell family—they were in fact a population of cells called DC3, which are known to promote inflammation.
Hence, instead of using CD14 as a marker for monocytes, the researchers recommended two alternative markers, CD88 and CD89, instead. On the other hand, they found that DC3 cells could be identified based on their expression of CD5, CD14 and CD163 markers.
Further analysis showed that DC3 cells play a significant role in systemic lupus erythematosus, an autoimmune disease more commonly diagnosed in women and characterized by a butterfly-shaped rash on the face. DC3 cells secreted several pro-inflammatory molecules such as IL-1α and CXCL1, which are known to contribute to lupus onset and progression.
Ginhoux noted that the InfinityFlow method used in this study will be useful for identifying immune cell types involved in other diseases such as cancer and atopic dermatitis. “From a clinical perspective, we may get to a point where we can pinpoint which immune cell is the ‘bad guy’ and develop drugs to kill or remove it,” Ginhoux said.
“We’re also preparing to release a paper detailing how to use InfinityFlow for profiling immune cells so that the technique is more accessible to the wider research community,” he concluded.
The A*STAR-affiliated researchers contributing to this research are from the Singapore Immunology Network (SIgN).