About

We are the Laboratory of Computational Systems Biology and Biophysics in the Department of Biomedical Engineering at Yale University. Explore the site to find out more about our research!

Research topics

• Artificial Intelligence
• Computer Science
• Mathematics
• Ecology
• Evolutionary biology
• Biology
• Physics
• Thermodynamics
• Classical mechanics
• Mathematical optimization
• Statistical physics

Selected publications

• Discovering interpretable low-dimensional dynamics using maximum entropy

  ArXiv.org · 2026-05-16

  articleOpen access

  Models (i.e., governing equations) are fundamental to science and engineering. Advances in data acquisition now make it possible to extract interpretable, low dimensional descriptions from high dimensional observations. However, existing approaches sacrifice either interpretability for reconstruction accuracy or infer symbolic dynamics without relating latent coordinates to physically meaningful observables. Here we present Edwin (maximum entropy driven compression with interpretable nonlinear model discovery), a unified framework that simultaneously performs dimensionality reduction using the dynamic maximum entropy (DME) principle and discovers sparse symbolic models governing latent dynamics, as well as the coupling between learned features and external metadata. We validate Edwin on diverse simulated systems, including stochastic diffusion, the Ornstein-Uhlenbeck process, self assembling particles, spiking neural populations, and low rank recurrent neural networks, as well as on a noisy experimental time series of aggregating RNA-liposome complexes. Across all systems, Edwin recovers low dimensional symbolic models that are physically interpretable and generalize to unseen conditions. Together, these results establish Edwin as a powerful framework for inferring interpretable, low dimensional dynamics directly from high dimensional data.

  PublisherOA PDF

• Discovering interpretable low-dimensional dynamics using maximum entropy

  arXiv (Cornell University) · 2026-05-16

  preprintOpen access

  Models (i.e., governing equations) are fundamental to science and engineering. Advances in data acquisition now make it possible to extract interpretable, low dimensional descriptions from high dimensional observations. However, existing approaches sacrifice either interpretability for reconstruction accuracy or infer symbolic dynamics without relating latent coordinates to physically meaningful observables. Here we present Edwin (maximum entropy driven compression with interpretable nonlinear model discovery), a unified framework that simultaneously performs dimensionality reduction using the dynamic maximum entropy (DME) principle and discovers sparse symbolic models governing latent dynamics, as well as the coupling between learned features and external metadata. We validate Edwin on diverse simulated systems, including stochastic diffusion, the Ornstein-Uhlenbeck process, self assembling particles, spiking neural populations, and low rank recurrent neural networks, as well as on a noisy experimental time series of aggregating RNA-liposome complexes. Across all systems, Edwin recovers low dimensional symbolic models that are physically interpretable and generalize to unseen conditions. Together, these results establish Edwin as a powerful framework for inferring interpretable, low dimensional dynamics directly from high dimensional data.

  PublisherDOI

• Endocytosis shapes extracellular chemical gradients in autonomous cell–cell attraction

  bioRxiv (Cold Spring Harbor Laboratory) · 2026-04-02

  articleOpen accessSenior authorCorresponding

  Receptor-mediated ligand endocytosis is traditionally viewed as a negative feedback mechanism for signal attenuation. Here we show that ligand removal can paradoxically enhance directional information in autonomous cell–cell attraction. Many cell systems migrate toward one another in the absence of externally imposed gradients, implying that secretion, diffusion, and uptake must themselves generate usable directional cues. We develop a surface-resolved theory of a finite-sized detector exposed to a nearby source and derive analytical expressions for the steady-state ligand field. The resulting concentration profiles are governed by a single dimensionless Damköhler number that compares receptor-mediated endocytosis to diffusive ligand transport. Increasing ligand removal lowers extracellular ligand concentrations and reduces absolute concentration differences across the detector surface, but preferentially enhances relative surface anisotropy. Thus, destroying the signal can increase the usable information encoded in relative gradients. Incorporating nonlinear downstream processing reveals a tradeoff between contrast enhancement and signal depletion that yields a well-defined optimal endocytosis rate, in a regime consistent with experimentally measured receptor internalization kinetics. These results recast receptor-mediated endocytosis as an extracellular information-processing mechanism that reshapes self-generated gradients to enhance directional information.

  PublisherOA PDFDOI

• Metagenome-scale Modeling to Assess Microbiome Metabolic Complementarity for Precision Microbiota Transplantation Therapies

  bioRxiv (Cold Spring Harbor Laboratory) · 2026-05-16

  articleOpen access

  Fecal microbiota transplantation (FMT) holds therapeutic promise beyond recurrent Clostridioides difficile infection, but clinical outcomes remain unpredictable, in part because existing computational models do not fully capture the metabolic compatibility between donor and recipient communities. Here, we present a metagenome-scale metabolic modeling framework that quantifies metabolic niche complementarity between donor and recipient microbiomes to predict transplantation outcomes. Using MICOM-derived community metabolic models, we show that donor taxa whose metabolic flux profiles are more dissimilar from the recipient community engraft at significantly higher rates in both murine and human FMT cohorts. In a human IBS trial, metabolic models accurately predicted post-FMT community composition via leave-one-out cross-validation and recapitulated disease-associated alterations in short-chain fatty acid, sulfur, and gas metabolism. We then performed 2,548 in silico FMT simulations between IBS-D/M patients and donors from the OpenBiome biobank to demonstrate a platform for personalized donor screening. This screen identified super-donors characterized by high taxonomic diversity, broad metabolic niche coverage, and community interaction networks dominated by cross-feeding rather than competition, as quantified by a flux-derived ecological network balance index that strongly predicted engraftment potential. This framework provides a mechanistic, scalable tool for rational donor-recipient matching that could guide personalized microbiome-based therapies.

  PublisherOA PDFDOI

• Non-equilibrium strategies for ligand specificity in signaling networks

  eLife · 2025-07-29

  preprintOpen accessSenior author

  Abstract Signaling networks often encounter multiple ligands and must respond selectively to generate appropriate, context-specific outcomes. At thermal equilibrium, ligand specificity is limited by the relative affinities of ligands for their receptors. Here, we present a non-equilibrium model showing how signaling networks can overcome thermodynamic constraints to preferentially signal from specific ligands while suppressing others. In our model, ligand-bound receptors undergo sequential phosphorylation, with progression restarted by ligand unbinding or receptor degradation. High-affinity complexes are kinetically sorted toward degradation-prone states, while low-affinity complexes are sorted towards inactivated states, both limiting signaling. As a result, network activity is maximized for ligands with intermediate affinities. This mechanism explains paradoxical experimental observations in receptor tyrosine kinase (RTK) signaling, including non-monotonic relationships between ligand affinity, kinase activity, and signaling output. Given the ubiquity of multi-site phosphorylation and ligand-induced degradation across signaling pathways, we propose that kinetic sorting provides a general non-equilibrium strategy for ligand discrimination in cellular networks.

  PublisherDOI

• Non-equilibrium strategies enabling ligand specificity by signaling receptors

  eLife · 2025-07-29

  articleOpen accessSenior author

  Signaling receptors often encounter multiple ligands and have been shown to respond selectively to generate appropriate, context-specific outcomes. At thermal equilibrium, ligand specificity is limited by the relative affinities of ligands for their receptors. Here, we present a non-equilibrium model in which receptors overcome thermodynamic constraints to preferentially signal from specific ligands while suppressing others. In our model, multi-site phosphorylation and active receptor degradation act in concert to regulate ligand specificity, with receptor degradation, a common motif in eukaryotes, providing a previously under-appreciated layer of control. Here, ligand-bound receptors undergo sequential phosphorylation, with progression restarted by ligand unbinding or receptor turnover. High-affinity complexes are kinetically sorted toward degradation-prone states, while low-affinity complexes are sorted toward inactivated states, both limiting signaling. As a result, network activity is maximized for ligands with intermediate affinities. This mechanism explains paradoxical experimental observations in receptor tyrosine kinase signaling, including non-monotonic dependence of signaling output on ligand affinity and kinase activity. Given the ubiquity of multi-site phosphorylation and ligand-induced degradation across signaling receptors, we propose that kinetic sorting may be a general non-equilibrium ligand-discrimination strategy used by multiple signaling receptors.

  PublisherDOI

• Minimax entropy: The statistical physics of optimal models

  ArXiv.org · 2025-05-02

  preprintOpen access

  When constructing models of the world, we aim for optimal compressions: models that include as few details as possible while remaining as accurate as possible. But which details -- or features measured in data -- should we choose to include in a model? Here, using the minimum description length principle, we show that the optimal features are the ones that produce the maximum entropy model with minimum entropy, thus yielding a minimax entropy principle. We review applications, which range from machine learning to optimal models of biological networks. Naive implementations, however, are limited to systems with small numbers of states and features. We therefore require new theoretical insights and computational techniques to construct optimal compressions of high-dimensional datasets arising in large-scale experiments.

  PublisherOA PDFDOI

• Emergent eukaryotic directional sensing via receptor degradation and diffusion

  Proceedings of the National Academy of Sciences · 2025-12-18

  articleOpen accessSenior authorCorresponding

  Directional sensing enables eukaryotic cells to detect spatial gradients of extracellular ligands, allowing them to orient and migrate within complex environments. Prevalent models explain this computation through a circuit where signaling species with distinct diffusion constants act incoherently on a downstream readout. Here, we propose a fundamentally different mechanism of directional sensing based on simple receptor-level processes. Our model integrates three ubiquitous receptor processes-lateral diffusion, basal ligand-independent activation, and active receptor degradation, THAT together synergize to generate robust directional sensing. In the absence of diffusion, active receptor degradation and basal activity implement an integral feedback that adapts active receptor levels to a ligand-independent set point while depleting receptors in relatively ligand-rich regions. Diffusion then redistributes receptors from regions of relatively low ligand exposure to regions with higher ligand exposure. This creates a spatial mismatch between activity and feedback that drives asymmetric receptor activity relative to the set point. The model predicts an optimal diffusion constant that maximizes polarization, reveals that receptors can encode relative rather than absolute ligand concentrations, and identifies the optimal basal activity that maximizes the signal-to-noise ratio in stochastic regimes. A survey of kinetic parameters across receptor families suggests that this diffusion-degradation synergy constitutes a broadly applicable, receptor-level mechanism for directional sensing.

  PublisherDOI

• Author response: Non-equilibrium strategies enabling ligand specificity by signaling receptors

  2025-10-29

  peer-reviewOpen accessSenior author
  PublisherDOI

• Author response: Non-equilibrium strategies enabling ligand specificity by signaling receptors

  2025-10-06

  peer-reviewOpen accessSenior author

  Signaling receptors often encounter multiple ligands and haven been shown to respond selectively to generate appropriate, context-specific outcomes. At thermal equilibrium, ligand specificity is limited by the relative affinities of ligands for their receptors. Here, we present a non-equilibrium model in which receptors overcome thermodynamic constraints to preferentially signal from specific ligands while suppressing others. In our model, multi-site phosphorylation and active receptor degradation act in concert to regulate ligand specificity, with receptor degradation providing a previously under-appreciated layer of control. Here, ligand-bound receptors undergo sequential phosphorylation, with progression restarted by ligand unbinding or receptor turnover. High-affinity complexes are kinetically sorted toward degradation-prone states, while low-affinity complexes are sorted towards inactivated states, both limiting signaling. As a result, network activity is maximized for ligands with intermediate affinities. This mechanism explains paradoxical experimental observations in receptor tyrosine kinase (RTK) signaling, including non-monotonic dependence of signaling output on ligand affinity and kinase activity. Given the ubiquity of multi-site phosphorylation and ligand-induced degradation across signaling receptors, we propose that kinetic sorting may be a general non-equilibrium ligand-discrimination strategy used by multiple signaling receptors.

  PublisherDOI

Recent grants

• MANET: Maximum Entropy Neural Networks for Mechanistic Modeling of Single Cell Behavior

  NIH · $1.9M · 2021–2026

Frequent coauthors

• Andrew Goetz

  Yale University

  25 shared

• Hoda Akl

  University of Florida

  20 shared

• D. Asthagiri

  16 shared

• Ken A. Dill

  Stony Brook University

  16 shared

• Brian W. Ji

  Brown University

  15 shared

• Dennis Vitkup

  Columbia University Irving Medical Center

  14 shared

• Mario Niepel

  10 shared

• Madan Krishnamurthy

  OPKO Health (Ireland)

  10 shared

Labs

• Computational Systems BiologyPI

Awards & honors

• Maximizing Investigator's Research Award, NIGMS 2022-2026