Menu Close

Overview

Research in our group uses theoretical approaches to gain insight into cellular mechanisms in vivo. Our main interests focus on development and regeneration, processes driven by diverse types of stem cells. These stem cell systems nevertheless share essential characteristics, such as the ability to specify and maintain cell fates in dynamic and uncertain environments. We address questions of cell dynamics, signaling, and fate control by mathematical modeling, statistical inference, computational method development, and data analysis. Below are examples of current research directions.


Single-cell analysis of cell state transitions


Top: Hematopoietic cell states identified by SoptSC (data from here). Bottom: cell-cell signaling interaction network (mediated by Tgf-β) and lineage relationships define cell state transitions during hematopoiesis.
Studying single cells in high resolution has led to significant advances, and in some cases has completely revised how we characterize and understand cell states. These data include single-cell/single-nucleus RNA-seq, single-cell chromatin accessibility (ATAC-seq), and spatial transcriptomics (e.g. MERFISH). Of particular importance in these data are those cell states which are not persistent and accompanied by well-defined functions, but instead occupy less well-characterized roles in a cell atlas, and in various contexts are referred to as intermediate, hybrid, or transition cell states. Single-cell profiling has advanced our ability to probe intermediate cell states, but new computational methods and theoretical models remain essential to explain and predict from these data. We develop computational tools to infer cell states transitions, cell-cell communication, and dynamic gene regulatory interactions that in part define or control cell fates.

Information about the signals that are passed between cells (via cell-cell communication) can be directly inferred from data, using e.g. expression of paracrine signaling factors, or can be inferred from models that explore different possible topologies of coupled gene regulation and cell-cell communication networks. By the integration of data with dynamical systems models, it becomes possible to perform parameter inference and model selection at the level of single cells.


Modeling stem cell lineage and signaling dynamics


Top: competitive dynamics of a stem cell lineage model with progeny- and niche-mediated feedback. Bottom: Analysis of the critical points determines regions (shaded) where healthy and leukemia stem cells coexist, and prognosis is most uncertain.
Individual stem cell decisions (self-renewal, differentiation, death) must be coordinated at many levels, from the molecular and cellular to the tissue/organismal, in order to maintain homeostasis. This is particularly important in high-turnover tissues such as blood, guts, and skin. In all of these systems, at multiple scales, complex and noisy nonlinear dynamics emerge. Using dynamical models of the interactions between cells and external signals from microenvironments/niches, we can learn the behavior of these systems and predict responses to perturbations (such as injury, infection, or cancer). An important perspective that we consider is the ecology of cells, i.e. the strategies that cells adopt in particular environments that can support either the interests of the organism (in the case of healthy multicellular life) or their own interests (in the case of cancer). Different life history strategies can lead to vastly different outcomes, and may explain, for example, some remarkable stem cell abilities such as homing to (and mobilization from) the niche, or competitive responses to niche invasion by cancer cells.

We perform parameter inference and model selection within a Bayesian framework to appropriately account for our knowledge and uncertainties about these systems, and to be able to predict cell dynamics under new perturbations. Analysis of mechanistic models, in light of data, gives insight into the tightly regulated transcriptional programs that underlie stem cell phenotype. For example, we can investigate how a balance of cell division, differentiation, and migration controls tissue development and regeneration.


Multiscale tissue dynamics

Top: Volume of niche space mapped out by hematopoietic stem cell movement from healthy (left), or T. spiralis-infected stem cells (right). Bottom: Ex vivo epithelial growth from the nephric duct branches during co-culture with signals from the mesenchyme (GDNF). We can reproduce such branching using a hybrid model of individual cell-based epithelial growth (black) and a continuous GDNF gradient (color).
Under conditions when cell noise is on the same order of magnitude as the dynamics, or when spatial cell distributions are complex and heterogeneous, differential equation-based modeling approaches are not sufficient to describe the biological systems of focus; we need agent-based/individual cell-based models to study the dynamics of such systems. Moreover, crosstalk at and between biological scales (that should not be ignored) complicates these dynamics, and as such new models and approaches are needed.

We develop individual cell-based models that target specific biological questions. For example, we seek to understand how stem cells interact with niches in 3D space, to predict what factors are crucial to initiate epithelial tissue branching, and to study interactions between the immune system and cancer. These models often contain hybrid components, for example continuous components (PDEs) to describe diffuse factors such as morphogens, and agent-based on/off lattice components to describe the agents (typically biological cells) that interact and are regulated both intrinsically and by the surrounding microenvironment.

Some examples of such multiscale models include kidney branching organogenesis and inflammation-induced tumorigenesis. Simulation times can become prohibitive for high-throughput analyses (e.g. parameter inference) of such models, and so we work on computational methods to improve simulation efficiency and to improve/facilitate Bayesian parameter inference. The overarching goal for each of these models is to be able to predict the emergent tissue-level behavior resulting from dynamic and noisy cell and molecular interaction networks.