Machine Learning-Enhanced Estimation of Cellular Protein Levels from Bright-Field Images.

In this study, we aimed to develop a novel method for non-invasively determining intracellular protein levels, which is essential for understanding cellular phenomena. This understanding hinges on insights into gene expression, cell morphology, dynamics, and intercellular interactions. Traditional cell analysis techniques, such as immunostaining, live imaging, next-generation sequencing, and single-cell analysis, despite rapid advancements, face challenges in comprehensively integrating gene and protein expression data with spatiotemporal information. Leveraging advances in machine learning for image analysis, we designed a new model to estimate cellular biomarker protein levels using a blend of phase-contrast and fluorescent immunostaining images of epidermal keratinocytes. By iterating this process across various proteins, our model can estimate multiple protein levels from a single phase-contrast image. Additionally, we developed a system for analyzing multiple protein expression levels alongside spatiotemporal data through live imaging and phase-contrast methods. Our study offers valuable tools for cell-based research and presents a new avenue for addressing molecular biological challenges.