Steven Reilly, PhD

The lab works across five inter-connected areas: we develop and apply advanced methods to read the non-coding genome, interpret its regulatory grammar, dissect the architecture of traits and disease, write novel function into regulatory elements, and trace the history of genetic adaptations across evolution.

Reading the non-coding genome. As a former ENCODE Functional Characterization Center, we build high-throughput experimental tools to directly characterize the non-coding genome, determining which regulatory elements are active in a cell, which genes they control, and how variation tunes their activity up or down, including saturation mutagenesis of individual elements. These approaches include large-scale non-coding CRISPR technologies such as HCR-FlowFISH, prime-editing technologies to perturb at single-base resolution, and associated analysis suites (CASA). We have also contributed to the development and application of massively parallel reporter assays (MPRAs), including their expansion to novel genomic functions such as 3′UTRs, and are now extending these readouts to single cells.

Interpreting regulatory grammar. We develop machine-learning models, such as MPAC, that predict base by base how a variant raises or lowers regulatory activity. We are particularly interested in approaches that scale to saturation mutagenesis of the entire regulatory genome, allowing us to decipher the rules governing cis-regulatory element function. We apply these tools to study the non-coding impacts of somatic mutations in cancer, rare disease-associated variants, the relationship between function and evolutionary constraint, and the molecular mechanisms underlying traits.

The genetic architecture of disease. We apply these tools in concert to dissect how inherited common and rare variants, together with somatic mosaic mutations, contribute to disease risk and human traits. Recent work has sought to dissect thousands of molecular and disease-associated traits, uncover causal somatic variants in schizophrenia, and apply single-cell CRISPR screens to chart the cis- and trans-regulatory networks connecting these variants.

Writing new genome function. We developed CODA, a deep-learning platform that designs synthetic regulatory elements with programmed cell-type specificity. We can now design enhancers and promoters with programmable activity for potential use in gene and cell therapies.

Selection and human evolution. Using our machine-learning model DeepSweep to pinpoint causal adaptive variants, together with our MPRA and CRISPR technologies, we characterize variants under natural selection in different human populations that may still contribute to modern human phenotypic differences. We draw on ancient DNA, trace the impact of archaic (Neanderthal and Denisovan) ancestry across the globe, and characterize the human-specific deletions (hCONDELs) that helped make us human.