Research Overview

The study of algorithms teaches us the limits of what we can learn about a system given access to certain data and operations. We are applying these lessons to develop new experimental methods, paired with computational approaches, that allow us to measure and perturb with far greater efficiency than is currently possible. We are applying these methods to understand tissue physiology at a molecular level, to explore cellular pathways and combinatorial genetic interactions, and to integrate each of these with the study of human genetics. Read about our perspective here.

Compressed Sensing

Composite measurements and compressed sensing

Our first, theoretical application described how to apply these ideas to make gene expression profiling more efficient by measuring a few random linear combinations of gene abundance.

Publication: Cleary et al, Cell (2017)


Composite measurements for imaging transcriptomics and proteomics

Building on the theoretical foundations above, we have developed a scheme, Composite In Situ Imaging (CISI), to make in situ measures of gene abundance exponentially more efficient. In CISI, each color in each round of imaging corresponds to a linear combination of genes. By measuring relatively few linear combinations, we can infer modular expression patterns in each region of a tissue section, then decompress to infer individual gene abundances and ultimately greatly increase throughput. We have demonstrated a 500-fold gain in throughput, and ultimately hope to make profiling large tissue volumes roughly 10,000x more efficient than it is today, enabling in situ studies of entire organs at single-cell resolution.

Preprint available on biorxiv.

Composite Measurements
Random composite perturbations


Random composite perturbations

We apply the same conceptual approaches to study the effects of genetic perturbations. Just as we don’t need to measure every gene — because many are co-regulated — we don’t need to individually perturb every gene, since many are co-functional, and their perturbations result in similar outcomes. We are thus developing methods of random composite perturbations, applied to Perturb- or CROP-Seq studies, and aim to use these method to elucidate regulatory circuits with ~1,000x greater efficiency. Ultimately, we hope to use random composite perturbations to make studies of combinatorial perturbations possible on a massive scale.

Group testing for SARS-CoV-2

Many of the approaches we take in the lab have their roots in group testing, going back to the 1950’s. We are studying and implementing combinatorial testing for SARS-CoV-2 positive sample identification, and pooled testing for prevalence estimation. This work has two goals: (i) to study pooled testing in the context of COVID epidemic dynamics and viral kinetics, and use these results to identify the most effective pooling strategies given different resource constraints; and (ii) to identify new combinatorial pooling designs that are maximally effective, simple to implement by hand, robust, and maximally balanced.

Preprint for (i) available on medrxiv.


Group testing