Current research
Biological histories contain the rules of change.
We recover histories, invent mathematics to learn what they mean, and perturb living systems to test whether those rules can control what happens next.
Biological history is rarely available as a movie. We make it observable by engineering cells to record their own histories, watching cellular families directly, or reconstructing the past from molecular traces.
A history alone is not an explanation. We develop mathematical frameworks that infer the hidden rules of change: which states cells inherit, how those states are regulated, and how populations expand.
We then perturb the system to test these rules and use what we learn to control cells, populations, and developing tissues.
01Synthetic biology · Lineage recording
Can cells record their own past?
Writing cellular history into DNA
One way to recover history is to make cells record it. We have developed engineered mouse models in which cells write their lineage histories into their genomes as they divide. The cells accumulate heritable DNA marks that are passed to their descendants. By sequencing these marks together with gene expression, we can reconstruct which cells shared an ancestor and what they became.
We are now building a new generation of recorders using base and prime editors. These systems will rewrite DNA without making double-strand breaks. They will record not only ancestry, but also which signals a cell experienced, in what order, and for how long.
These records are necessarily incomplete and noisy. We therefore develop probabilistic models that reconstruct lineage and cell-state transitions from partial histories. Our goal is to discover which histories produce particular cell fates and then recreate those histories in the laboratory.
Heritable DNA marks written as cells divide and receive signals.
Probabilistic reconstruction of partial, noisy lineage and signaling histories.
Replay the histories that produce desired cell fates.
03Cancer genomics · Population genetics
Can present-day blood reveal when cancer began?
Reconstructing the origins of blood cancer
Blood cancers can begin decades before symptoms appear. We discovered this by reconstructing the lineage trees of individual cancer cells from naturally occurring somatic mutations in their genomes.
Each time a blood stem cell divides, it can acquire random mutations and pass them to its descendants. Cells that share mutations share ancestry. By comparing their genomes, we recovered the history of the cancer and inferred when the cancer-causing mutation first arose and how rapidly the mutant clone expanded.
We are now developing mathematics that can recover this history from a single present-day blood sample. The distribution of mutation frequencies—called the site-frequency spectrum—contains a compressed record of the population’s past. We use this spectrum to infer the age and fitness of mutant clones and ask whether the past trajectory of a cancer can predict its future course.
Natural somatic mutations that act as lineage marks in blood cells.
The age and fitness of mutant clones from trees and mutation-frequency spectra.
Use past trajectories to identify cancers at risk of future progression.
Development · Synthetic embryos
Can we recreate and control human development?
Recreating human development in a dish
Development shows why history matters. An embryo does not assemble from a blueprint imposed from outside. Cells divide, signal to their neighbors, and choose fates. Their collective decisions create a body plan.
We create synthetic embryos from human pluripotent stem cells that recapitulate key events of embryonic development in a dish. These systems undergo symmetry breaking, form a body axis, and produce early tissue patterns. They allow us to observe and perturb developmental histories that cannot be studied directly in humans.
We combine lineage tracing, DNA recorders, live reporters, and single-cell sequencing to determine which sequence of states and signals produces each fate. Rather than adding signals according to a fixed schedule, we want cultures that sense their own developmental state and adjust signals in real time. Our long-term goal is to generate functional blood stem cells.
Lineage, signaling, and cell-state histories during self-organization.
Which sequences of founder states and signals generate each tissue fate.
Cultures that sense developmental state and adjust signals in real time.
05Single-cell genomics · Bacteria
What is a cell state in a bacterium?
Decoding bacterial cell states
In bacteria, history is often stored as cellular state. Genetically identical cells can behave differently because they have experienced different environments and activated different regulatory programs. These states can determine whether a cell grows, persists, becomes pathogenic, or survives an antibiotic.
We developed proBac-seq, a probe-based method for measuring gene expression in individual bacteria. We are now developing perturBac-seq, which measures the transcriptional state of each cell together with the CRISPR perturbation it received. This will allow us to test thousands of genetic perturbations in parallel.
The computational question is equally fundamental: what constitutes a cell state in a bacterium? We develop methods that find meaningful structure in high-dimensional expression data and determine how genetic and environmental perturbations move cells between states.
Gene expression and CRISPR perturbations in individual bacterial cells.
Meaningful states and causal regulatory programs in high-dimensional data.
Redirect bacteria away from pathogenic, persistent, or drug-resistant states.
06Experimental evolution · HIV
Can we predict evolution from viral genomes?
Watching evolution unfold in viruses
At longer timescales, history becomes evolution. Mutations accumulate in genomes and record the paths that populations have taken. But conventional sequencing often separates these mutations, making it difficult to determine which combinations occurred in the same organism.
We use HIV as a controlled experimental system for watching evolution unfold. We are developing methods to recover full-length genomes from individual viruses and follow viral populations as they evolve in the laboratory. Complete genomes reveal which mutations travel together and how interactions between mutations affect reproductive success.
We develop mathematical models that use these sequences to infer fitness landscapes: maps connecting viral genomes to their ability to reproduce. We then ask whether these landscapes can predict which evolutionary paths a population will take and use what we learn to engineer better viral systems for gene delivery.
Full-length genomes from individual viruses evolving in the laboratory.
Fitness landscapes connecting combinations of mutations to reproductive success.
Predict evolutionary paths and engineer better systems for gene delivery.
A theory that runs through every project
Learning the rules that connect history to fate
Every project in the lab meets at the same question: can we turn a biological history into a predictive theory? DNA marks written in a mouse, the behavior of related cells, the mutation spectrum of a cancer, the states of individual bacteria, the shape of a developing embryo, and the genomes of evolving viruses are all different kinds of evidence about processes that unfolded through time.
High-throughput experiments now give us enough evidence to build models at a scale that was once impossible. These models can contain thousands of cells, genes, interactions, and unknown parameters. We cannot fit each parameter by hand. Instead, we write the model as a differentiable computer program. Automatic differentiation calculates how every parameter would change the model’s prediction, and gradient descent adjusts them together until the model accounts for the data.
We did this for development. Starting with independent cells, we trained a model to learn the local rules of movement and signaling that cause those cells to assemble into a chosen three-dimensional form. We want to carry the same idea across the lab: learn hidden rules from biological histories, test those rules with perturbations, and discover which rules must change to produce a different fate.
Present-day cell states together with lineage, imaging, and perturbation data.
Large differentiable models fit with automatic differentiation and gradient descent.
Learn which local rules must change to produce a different biological future.
Mathematics across the lab
Join the work
Find your question.
These projects are designed for students who want to move between experiment and theory.
Explore rotations