Research

We develop computational methods to connect gene regulation, cellular states, and tissue organization with human disease.

Human disease emerges through changes in gene regulation, cellular states, and interactions within tissues. Our research asks how these molecular and cellular processes can be reconstructed from genomic measurements, and how genetic variation and perturbations reshape them. Single-cell and spatial omics make it possible to resolve these processes in their biological context, but the data are sparse, heterogeneous, and incomplete. We develop computational models that connect genomic sequence, chromatin accessibility, gene expression, and tissue organization while making those limitations explicit.

Our approach combines AI method development with large-scale human genomics. We build models of regulatory networks and cellular dynamics, integrate measurements across assays and samples, and use computational perturbations to generate testable hypotheses. Work in ENCODE, PsychENCODE, and SCORCH connects these methods to shared genomic resources and disease-focused collaborations. Across the four programs below, we aim to identify which molecular programs are shared across conditions, which are specific to a cell type or tissue context, and which regulatory changes warrant experimental follow-up for therapeutic discovery.

AI & Machine Learning for Single-Cell and Spatial Genomics

Scientific Question / Motivation

Single-cell and spatial measurements reveal cellular heterogeneity, but sparsity, incomplete sampling, and differences between assays complicate the reconstruction of cell states and tissue organization.

Our Approach

We develop machine-learning models that combine gene expression, cellular relationships, and spatial context. Our work addresses continuous cell-state dynamics, missing expression measurements, and the integration of tissue sections, with an emphasis on representations that retain biological structure.

Current Direction

We aim to connect models of cellular dynamics with tissue context and extend integration across samples and modalities. A central question is how to preserve meaningful biological variation while accounting for technical differences and uncertainty in reconstructed measurements.

Representative Work

Gene Regulation & Functional Genomics

Scientific Question / Motivation

Gene regulation connects genomic sequence and chromatin organization to cell-specific expression. Understanding this process requires identifying regulatory elements, their target genes, and the context in which regulatory relationships change.

Our Approach

We integrate functional and regulatory genomics with single-cell measurements to study the non-coding genome and infer regulatory networks. These models connect molecular features across scales, from cis-regulatory elements to interactions between cells.

Current Direction

Our direction is to connect sequence-level predictions, regulatory networks, and perturbation evidence. We aim to distinguish context-specific regulatory effects from shared programs and develop hypotheses that can be assessed with independent functional measurements.

Representative Work

Human Disease Genomics

Scientific Question / Motivation

Disease-associated genetic variation and molecular changes act within particular cell types and tissue environments. Resolving these contexts is necessary to move from statistical associations toward mechanistic hypotheses.

Our Approach

We study disease-associated cellular states and regulatory programs by integrating single-cell, spatial, and functional genomics. The lab’s scope includes psychiatric, neurodegenerative, substance-use, HIV-associated, and other complex human diseases.

Current Direction

We aim to connect disease-associated programs to their cellular and spatial contexts, and to assess which relationships generalize across individuals and disease settings. Integrating these observations with perturbation evidence is a route toward more specific, testable disease mechanisms.

Representative Work

AI-Guided Perturbation & Therapeutic Discovery

Scientific Question / Motivation

Genes and regulatory elements act in combinations, making exhaustive experimental perturbation difficult. Computational models can help prioritize candidate interventions and identify which combinations warrant experimental evaluation.

Our Approach

We develop interpretable models of regulatory dependencies and use in silico perturbations to study how molecular changes may alter cell states. This work connects disease-associated programs with the prioritization of regulatory genes and candidate therapeutic hypotheses.

Current Direction

We aim to link regulatory perturbations to disease-relevant cellular outcomes and improve the selection of candidates for experimental follow-up. Model predictions provide hypotheses for validation; establishing therapeutic effects requires direct experimental evidence.

Representative Work