From observing to predicting single-cell structure and function with high-throughput/high-content microscopy

Abstract

In the past 15 years, cell-based microscopy has evolved its focus from observing cell function to aiming to predict it. In particular-powered by breakthroughs in computer vision, large-scale image analysis and machine learning-high-throughput and high-content microscopy imaging have enabled to uniquely harness single-cell information to systematically discover and annotate genes and regulatory pathways, uncover systems-level interactions and causal links between cellular processes, and begin to clarify and predict causal cellular behaviour and decision making. Here we review these developments, discuss emerging trends in the field, and describe how single-cell 'omics and single-cell microscopy are imminently in an intersecting trajectory. The marriage of these two fields will make possible an unprecedented understanding of cell and tissue behaviour and function.

Keywords: Machine Learning; causal cell behaviour; gene regulatory networks; genome-wide screening; high-content microscopy; high-throughput microscopy.

Figure 1. The principle of high-throughput/high-content microscopy screens to identify and study genes’ functions at scale

(A) Typical high-throughput/high-content microscopy pipeline, where the inputs are cells and cell images, and the output is high-dimensional, single-cell feature data. (B) Implementation steps involved in the context of a single perturbation (KD/KO) genome-wide phenotypic screen, allowing to systematically discover gene functions at scale.

Figure 2. Reconstructing gene/protein networks and systems-level interactions between cellular processes

Using two interventions, either by double gene KD/KO (A) or by combining gene KD/KO with fluorescent protein (FP) tagging (B), allows the reconstruction of functional interactions between genes/proteins and construction of regulatory networks.

Figure 3. Predicting spatio-temporal cellular structure using Deep Learning

The three main current uses of Deep Learning are: (A) phenotypically labelled images of whole image fields or single cells are used to learn a classification and/or a set of features for future uses; (B) pixel-level segmentation masks are used to train pixel-to-pixel networks for automated classification; (C) generative networks are given pairs of multi-channel images and trained to generate one channel given the others.

Figure 4. Measuring and predicting causal cell behaviour by time-resolved, continuous single-cell microscopy

(A) Single-cell microscopy (represented by a multi-generation cell lineage) allows tracking in real time spatiotemporal cell decisions – e.g. how cells grow, divide, die, migrate or change their structure, function, location and fate within a population – and the cell-to-cell heterogeneity of those decisions in ways inaccessible and complementary to (B) single-cell ‘omics (represented by a pseudo-time ordered PCA plot of ‘omics data from a cell population; each cell is 1 dot). As explained in the text this is epitomised by the time and location of cell death events (marked with an ‘X’ in (A)), which though crucial to cell population and tissue evolution are invisible to single-cell ‘omics approaches but visible by single-cell microscopy.