APDC: An accessible, open-source package for smart automation of single-cell fluorescence microscopy experiments

Single-cell fluorescence microscopy experiments provide highly quantitative information about spatiotemporal dynamics and heterogeneity of gene expression, but two major practical challenges limit their utility. One is their high tunability, with vastly many potential designs, e.g. combinations of labeled genes, induction levels, and expression measurement times. Another is the considerable hands-on experimenter time often required for data collection, mostly spent finding cells of interest and suitable imaging conditions.

Our open-source software package APDC (Acquire-Process-Decide-Control) automates design and execution of such experiments, extending the current approach of microscopy event specification across “multi-dimensional axes” to express experiment (re)designs in a control-theoretic framework, interconnecting operations to acquire and process images, make decisions, and control experimental conditions while integrating sequential model-based design of experiments (MBDoE) to accelerate discovery by selecting optimal numbers of cells to image at each experimental condition. Multiple optimality criteria allow experimenters to minimize total or largest uncertainty or to maximize the precision of particular subsets of parameters. APDC implements these functionalities for simulated as well as physical microscopes, allowing for orthogonal method prototyping and testing independently of the particulars of any real-world apparatus.

We validate these innovations by studying the glucocorticoid receptor, whose intracellular localization regulates the expression of dual-specificity phosphatase 1, part of signaling cascades involved in diverse cellular processes and whose dysregulation therefore contributes to the onset and evolution of numerous diseases. Investigations are conducted across multiple cell lines, with further acceleration via MBDoE, utilizing kinetic parameters fit to data from one line to serve as a Bayesian prior for another. We demonstrate dramatically faster and cheaper experiments without loss of accuracy, reducing hands-on experimenter microscopy time and accelerating research projects by an order of magnitude.