Modeling human red blood cell damage using experiments and machine learning

Circulating human red blood cells (erythrocytes) are exposed to fluid stressors. If the stressors are large enough, these erythrocytes can be damaged, and even destroyed (hemolysis). In particular, hemolysis can cause organ injury, negatively affecting a person’s health. While prior studies have shown an increase in hemolysis with the fluid shear rate, they contain large uncertainties in the experimental results, too large to offer insight on a per-person basis. Here, we predict hemolysis of human erythrocytes using a combination of experimental measurements and machine learning (ML). We conduct microfluidic experiments to directly measure the degree of hemolysis after the erythrocytes are exposed to a known stressor. We then combine experimental measurements, patient demographics, and ML-modeling. Additionally, we demonstrate the use of active learning to guide experiment selection. We anticipate these results to offer insights on how to combine human biology-focused experiments and ML. Specifically, these results may drive the development of new devices and procedures designed to reduce hemolysis on a per-patient basis.