| dc.description.abstract | Phenotypic diversity is a common feature shared between all cell populations, ranging from microbial cultures to multicellular organisms. It is a consequence of the stochastic nature of gene expression and regulation, which has been extensively studied over the years. Historically, the investigation of cell populations has been restricted to bulk as- says, limiting the information for cellular individuality. Over the past two decades, new technologies have been developed for studying single-cells at the genome, transcriptome, proteome, and metabolome level ranging from high-throughput to whole-genome sequencing of a few single cells. Yet, technical challenges of such technologies still exist, including handling of low volume materials, isolating single cells, and/or laborious labelling with fluorescence probes which could harm cells or interfere with biological targets. Microfluidics, and in particular droplet microfluidics, have shown great potential for tackling such issues due to the high level of control, high throughput, and low reagent costs. In addition, recent developments in deep neural networks have shown great potential for multi-class detection and classification for a variety of application, ranging from human facial recognition to disease diagnosis. Using a combination of such tools, along with microscopy, we first demonstrate the potential of microfluidics by developing a platform for algal cell encapsulation in stationary droplets to investigate the motility of two algae species. We analyze the control network underlying the transition of motility states and perform on-demand fusion of droplet pairs containing single cells and a chemical to understand how microorganisms adapt to sudden changes in their environment. Building upon this work, our objective was to enhance the enrichment of single cells during entrapment and establish an efficient pairing mechanism for the droplets. To achieve this, we report a custom-made convolutional neural network (CNN) architecture and explore the limits of detection by showcasing methods for CNN optimization, including artificial image augmentations for algorithm training. Using this approach, we successfully sort a variety of biological objects such as polystyrene spheres (PS), polyacrylamide beads (PA), breast cancer cells, 3D spheroids, and algae cells. Furthermore, the exploration of multiple object encapsulation and sorting was required to demonstrate novel droplet-based protocols requiring single cells co-encapsulated with single-beads, acting as unique cell identifiers.
We use the same microfluidic sorting device and implement a novel algorithm called YOLO (You Only Look Once) to increase the sorting through- put and accuracy, as well as perform multi-object sorting of mammalian cells and PA beads. To demonstrate applicability of this new type of object detector, we developed a method for investigating spatial association and dynamic interactions of cell doublets, which can be valuable for cell biology research, understanding disease mechanisms, drug discovery, biomedical engineering, and cancer research. Our findings and methodologies have the potential to advance various disciplines and provide valuable insights into the behavior and characteristics of biological systems at the single-cell level. | en_GB |
