Thomas Norman

CRISPR genetics and the explosion of experimental methods with single-cell resolution have each transformed our ability to study cell state and how it is perturbed in disease. A second revolution is now unfolding following the marriage of these two fields to produce single-cell CRISPR screens, in which many genetic perturbations (e.g. gene inhibition/activation by CRISPRi/a) are linked in pooled format to high-content measures of their effects (e.g. the transcriptome or morphology). These rich data can both identify genetic perturbations that elicit a particular behavior and catalog the spectrum of phenotypes associated with each genetic perturbation—i.e., one can perform “forward” and “reverse” genetics within the same experiment. The Norman Lab played a foundational role in the development of one of these techniques, Perturb-seq, and continues to develop new approaches to enable massive functional genomics experiments.

Figure 1.

A central hypothesis in the lab is that as these single-cell functional genomics approaches scale they will evolve to become tools for directed engineering of cell state. For example, given a cell type and a set of relevant genes, is it possible to predict combinations of those genes whose overexpression will steer the cells towards a desired state? Or can we at least narrow the field so that fewer combinations need to be tested experimentally? This type of directed approach to studying how genes influence cellular phenotype would have many applications. One concrete objective would be to better reconstitute the many in vivo states that ongoing cell atlas projects will reveal, enabling easier and more scalable experiments on in vitro facsimiles. More ambitiously, such experiments could inform efforts to therapeutically normalize the corresponding in vivo cell state. Finally, many fundamental problems in genetics are simply too large to study exhaustively even in the most tractable systems, making a directed approach the only option.

The key to making the transition from discovery to engineering of cell states is the addition of a predictive model—some notion of how distinct perturbations interact with each other to yield new phenotypes. We believe techniques like Perturb-seq can provide datasets of the scale, richness, and resolution needed to start building computational models of how phenotypes emerge and that serve as a natural complement to modern machine learning methods that thrive on large interventional datasets.