UC Irvine

Traditionally, phenotyping cells has been approached through the biological lens of looking at proteins, DNA, RNA, or other molecular descriptions of cells. We present the use of physical properties of cells that emerge as as a consequence of these molecular properties, as a means to mechanically phenotype single cells. In this work, we phenotype two cell types, mammalian and bacterial cells, through separate approaches. In the first study, we create a high-throughput single cell microfluidic stretcher, where we track non-linear dynamics of deforming cells at a high spatial and temporal resolution. We apply image and sequence-based deep learning models for feature extraction that are trained using time sequences of cell shapes, and we found to achieve high classification accuracy based on cytoskeletal properties alone. The best model classified sub-populations with an accuracy over 90%, significantly higher than the 75% we achieved with traditional methods. This increase in accuracy corresponds to a five fold increase in potential enrichment of a sample for a target population. In the second study, we utilize an electrical impedance based approach, called resistive-pulse sensing, to classify bacteria cells based upon their size, shape, and surface potentials. In this method, as individual bacteria cells transit the polymer pore, we observe a changes in electrical impedance proportional to the particle volume, and differences in transit time that is proportional to the surface potential. These features are used to classify three bacteria populations which differ in shape at a high accuracy through unsupervised clustering. We compare this approach to a shape based approach, upon which the resistive-pulse signal is averaged by bacteria of different lengths as they travel along an undulating micropore. This work establishes the application of sequence-based machine and deep learning models to dynamic deformability cytometry and resistive-pulse sensing of bacteria.