Skip to Content

Coronavirus: Get complete details about the university's response to COVID-19.

Department of Mathematics

  • Scatter Plot showing the model’s ability to distinguish between agonists inducing NETosis

Collaboration between UofSC, UNC, UTHSC Uses Machine Learning to Inform Fundamental Cell Biology

Image: Class Probability Plot showing the model's ability to distinguish between agonists inducing NETosis, with probability plotted against sample number.

NETosis is a fundamental process involving immune function and blood clotting, only recently discovered, where leukocytes – the most abundant type of white blood cell – unravel their DNA and extrude it as a physical and chemical net into surrounding plasma.  Scientists at UNC and UTHSC studying NETosis have faced a major hurdle: in order to determine the underlying biochemical processes, microscope images showing hundreds of thousands of cells treated in various ways must be analyzed.  However, even just counting the number of cells of various phenotypes (whether they have undergone NETosis or another process, for example) is extremely laborious and error-prone.

Cell classification
Cell classification by phenotype, pixel-level model.

 

With researchers in UofSC Computer Science and UNC, senior author Prof. Joshua Cooper of UofSC Mathematics has shown that not only is it possible to automate this process to provide invaluable data for biologists using convolutional neural networks (CNNs) and other machine learning techniques, but also that the resulting model can

  1. be more accurate than humans despite being trained on data annotated by humans,
  2. be deployed on inexpensive commodity hardware,
  3. differentiate between previously indistinguishable NETosis mechanisms,
  4. uncover systemic problems in prior experiments, and
  5. transform qualitative features of cells, such as their clustering tendencies or nuclear morphology, into measurable quantities that provide key biological insights.

These methods can be used in a wide range of cell microscopy settings, and the researchers anticipate that many more applications will arise in basic science for the augmentation of human researchers’ faculties with new machine learning instruments.

Their publication describing this highly multidisciplinary research funded by the NIH-NIAID, Doris Duke Foundation, and IBM has appeared recently in Nature Scientific Reports, demonstrating the extraordinary capabilities of modern mathematical ideas to transform research methods throughout the sciences.


Challenge the conventional. Create the exceptional. No Limits.

©