High-throughput Phenotyping of iPSC-derived Airway Epithelium by Multiscale Machine Learning Microscopy

PROJECT SUMMARY/ABSTRACT Challenges. The airway epithelium consists of various cell types – understanding cellular and functional heterogeneity will have a significant impact on diagnosing and treating diseases. However, few analytical tools are available to investigate spatiotemporal phenotypes of these cells on a global population scale. Conventional high-throughput microscopy (HTM), although powerful for dissecting heterogeneous biological processes, is significantly limited in multiscale imaging and analytics. Most HTM systems are constructed by combining high-magnification microscopes with scanning stages; this configuration would entail high complexity in the system design and operation, high cost, and slow image acquisition rates. Follow-on data analyses, based on traditional ensemble averaging approaches, often lead to the loss of detailed mechanistic information. Innovations. We will advance a “smart” imaging platform, M3 (Multiscale Machine-learning Microscopy) for large-scale, live-cell analyses. M3 will integrate cutting-edge breakthroughs: Fourier ptychographic microscopy (FPM) and deep learning (DL). FPM is based on a spatially coded-illumination technique, collecting low-resolution image sequences while changing the position of a point-light source. These images are then numerically combined to restore the whole Fourier space, allowing FPM to achieve both wide field-of-view and high spatial resolution simultaneously. DL is potent in discovering intricate, hidden structures in high-dimensional data sets with limited human supervision. We will integrate DL with time-series modeling to learn disease-related cellular traits. Goals. We will implement the M3 platform and adopt it to analyze cellular phenotypes during airway epithelium development. Aim 1. We will construct the M3 imaging system based on the FPM technology. This system will feature i) a new numerical algorithm to reconstruct 3D volumetric images and ii) multi-color imaging capacity for molecular detection. Aim 2. We will advance a DL framework for M3 image analyses. This framework will be designed to recognize different cell types and learn their spatiotemporal features to unravel multiscale cellular heterogeneity. Aim 3. We will apply M3 to phenotype cells in the airway epithelium. We will use an in-vitro model that uses induced pluripotent stem cells (iPSCs) to derive lung epithelium. M3 will monitor cellular differentiation during epithelium development and examine the correlation between cellular phenotypes and functionals. Impact. The M3 will bring unprecedented analytical power to characterize diverse cells within the airway epithelium, allowing us to discover many hidden phenotypes in cellular and tissue levels. Such knowledge would have implications for early disease detection as well as designing effective therapeutics.