Learning cytometers and beyond

Conventionally, cytometry has been a tool used based on human hypotheses. Typically, we use a biomarker to label cells and measure the intensities of markers and other representative values for defining the cells to be sorted. This approach is often described as "finding a needle in a haystack." In other words, we are actually dismissing cells as hays because they are difficult to label.
I started thinking about this about 10 years ago. What if we don't have good markers to define the cells of interest? What if we don't know which cells to label? What if we don't know how to recognize image patterns in the case of high-content cytometry? My idea was to reverse this line of thinking and develop a data-driven learning cytometry technology. In today's talk, I will introduce several of these learning cytometry techniques.
First, I will talk about ghost cytometry (GC). GC can be used to classify cell types, states, differentiation, etc. (as long as the morphological information is discernible) and selectively sort cells with high throughput. I show you GC-based machines that integrate various data analysis methods to robustly discriminate cells without markers. I also hope to introduce further new technologies, such as very sensitive and high throughput exosome analysis technology, as well as a label-free bacterial cell analyzer, which is expected to be applied to sepsis research.