Deciphering the biophysical drivers of interactions What decides which molecules bind which

We want to decipher the biophysical drivers of molecular interactions. Protein–protein interactions are one example: with the Dickinson lab, we use PANCS selections of designed binders against many targets to ask whether a sequence-only model can learn the rules for which proteins bind which, and to map the fitness landscapes of protein–protein interactions.

Research question

  • Protein–protein interactions. Can a sequence-only model learn the rules for which proteins bind which? (Aditya Singh, with the Dickinson lab)

Papers

  1. Mapping the diverse topologies of protein–protein interaction fitness landscapes
    S. S. Lu†, M. J. Styles†, C. F. Gao†, A. Nandy, C. Basile, J. A. Pixley, S. Tao, A. R. Dinner, S. Vaikuntanathan*, B. C. Dickinson*
    bioRxiv (2025) · submitted · †equal contribution · *joint corresponding authors bioRxiv

Cell states, decisions and control Decision making in high dimensions

How do cells make decisions from high-dimensional inputs, when receptors read combinations of ligands and a population of cells responds to a perturbation across thousands of genes? One line of our work asks this question using non-equilibrium network models. Response constraints on the steady states of these networks enable local, imperfect feedback control; the same constraints make each output monotone in each input, which limits the decision boundaries a network can draw. We are now asking which structural features of signaling networks let them draw complex decision boundaries. With the Goyal lab, we showed that the gene–gene covariance of unperturbed cells predicts transcriptome-wide responses to genetic and chemical perturbations (CIPHER).

Research questions

  • Decision boundaries and control. Which structural features let a biochemical network draw complex decision boundaries? (Carlos Floyd)
  • Perturbation responses. Do the fluctuations of unperturbed cells already predict how they respond to perturbation? (Vidyesh Rao Anisetti and Hector Lopez Rios, with the Goyal lab)

Papers

  1. Learning to control non-equilibrium dynamics using local imperfect gradients
    C. Floyd, A. R. Dinner, S. Vaikuntanathan
    Newton 2, 100595 (2026) Paper arXiv
  2. Limits on the computational expressivity of non-equilibrium biophysical processes
    C. Floyd, A. R. Dinner, A. Murugan, S. Vaikuntanathan
    Nature Communications 16, 7184 (2025) Paper
  3. Local imperfect feedback control in nonequilibrium biophysical systems enabled by response constraints
    C. Floyd, A. R. Dinner, S. Vaikuntanathan
    Physical Review X, in press (2026) arXiv
  4. Fluctuation structure predicts genome-wide perturbation outcomes
    B. Kuznets-Speck, L. Schwartz, H. Sun, M. E. Melzer, N. Kumari, B. Haley, E. Prashnani, S. Vaikuntanathan, Y. Goyal
    Nature Biotechnology, in press (2026) bioRxiv
Cellular recognition as classification: ligand–receptor affinity structure in immune recognition and glycan-based communication, the signaling cascade that computes the response, and low and high path multiplicity (Floyd et al., in preparation).
Cellular recognition as classification: ligand–receptor affinity structure in immune recognition and glycan-based communication, the signaling cascade that computes the response, and low and high path multiplicity (Floyd et al., in preparation).
A thermodynamic constraint on non-equilibrium response enables control of Markov jump processes by local feedback connections (Floyd, Dinner and Vaikuntanathan, Physical Review X, in press).
A thermodynamic constraint on non-equilibrium response enables control of Markov jump processes by local feedback connections (Floyd, Dinner and Vaikuntanathan, Physical Review X, in press).

Chemical drivers of flexible computing in brains What neuropeptides do for computation in the brain

In C. elegans, neuropeptides act far beyond the synaptic connectome and are needed to sustain behavioral states. We ask what neuropeptides do for computation in the brain. With Monika Scholz, we model their action as a slow, activity-dependent, multiplicative modulation of each neuron. With the synaptic couplings unchanged, this modulation extends memory retrieval far beyond the classical Hopfield capacity and produces emergent multistability.

Research question

  • What do neuropeptides add to what a neural network can store and compute? (Hector Lopez Rios and Daiki Goto, with Monika Scholz)

Papers

  1. Neuromodulation-inspired gated associative memory networks: Extended memory retrieval and emergent multistability
    D. Goto, H. M. Lopez Rios, M. Scholz, S. Vaikuntanathan
    Proc. Natl. Acad. Sci. USA, in press (2026) arXiv
Neuromodulator networks of C. elegans (serotonin, dopamine and three neuropeptide–receptor pairs) overlaid on the synaptic connectome, and the model, in which a neuromodulator layer z acts on the neural layer x through neuronal activity. Network data from Ripoll-Sánchez et al. (Neuron 2023) and Cook et al. (Nature 2019) (Goto et al., PNAS, in press).
Neuromodulator networks of C. elegans (serotonin, dopamine and three neuropeptide–receptor pairs) overlaid on the synaptic connectome, and the model, in which a neuromodulator layer z acts on the neural layer x through neuronal activity. Network data from Ripoll-Sánchez et al. (Neuron 2023) and Cook et al. (Nature 2019) (Goto et al., PNAS, in press).
Steady-state overlap against memory load α and initial overlap m(0), without (top) and with (bottom) neuropeptide modulation, and their difference (right); the dashed line marks the Hopfield capacity (Goto et al., PNAS, in press).
Steady-state overlap against memory load α and initial overlap m(0), without (top) and with (bottom) neuropeptide modulation, and their difference (right); the dashed line marks the Hopfield capacity (Goto et al., PNAS, in press).

Cytoskeletal dynamics Learning and adaptation in cytoskeletal networks

We study how cytoskeletal networks, the structural scaffolding of cells, learn and adapt through mechanical feedback and active processes. Our work shows how activity enables transitions between contractile and extensile states, and how mechanosensitivity enables a form of physical learning.

People: Lissa Rennert and Billie Meadowcroft

Papers

  1. Contractile to extensile transitions and mechanical adaptability enabled by activity in cytoskeletal structures
    A. Lamtyugina, D. S. Banerjee, Y. Qiu, S. Vaikuntanathan
    Biophysical Journal, in press (2026) Paper
  2. Diffusion models learn underlying trends in actomyosin networks and predict behavior at unseen filament turnover
    E. Rennert, A. K. Behera, Y. Qiu, S. Vaikuntanathan
    bioRxiv (2026) · submitted to Biophysical Journal bioRxiv
  3. Learning via mechanosensitivity and activity in cytoskeletal networks
    D. S. Banerjee, M. J. Falk, M. L. Gardel, A. M. Walczak, T. Mora, S. Vaikuntanathan
    PRX Life 4, 013001 (2026) Paper arXiv
  4. A thermodynamic framework for nonequilibrium self-assembly and force–morphology tradeoffs in branched actin networks
    E. Rennert, S. Vaikuntanathan
    eLife 14, RP104513 (2025) Paper eLife
  5. A unified model for the dynamics of ATP-independent ultrafast contraction
    C. Floyd, A. T. Molines, X. Lei, J. E. Honts, F. Chang, M. W. Elting*, S. Vaikuntanathan*, A. R. Dinner*, M. S. Bhamla*
    Proc. Natl. Acad. Sci. USA 120, e2217737120 (2023) · *joint corresponding authors Paper
  6. Elucidating the role of filament turnover in cortical flow using simulations and representation learning
    Y. Qiu, E. D. White, E. M. Munro, S. Vaikuntanathan*, A. R. Dinner*
    arXiv (2023) · under revision, Proc. Natl. Acad. Sci. USA · *joint corresponding authors arXiv
  7. Thermodynamic control of activity patterns in cytoskeletal networks
    A. Lamtyugina, Y. Qiu, E. Fodor, A. R. Dinner, S. Vaikuntanathan
    Phys. Rev. Lett. 129, 128002 (2022) Paper
  8. A strong nonequilibrium bound for sorting of cross-linkers on growing biopolymers
    Y. Qiu, M. Nguyen, G. M. Hocky, A. R. Dinner, S. Vaikuntanathan
    Proc. Natl. Acad. Sci. USA 118, e2102881118 (2021) Paper
  9. Liquid behavior of cross-linked actin bundles
    K. L. Weirich, S. Banerjee, K. Dasbiswas, T. A. Witten, S. Vaikuntanathan, M. L. Gardel
    Proc. Natl. Acad. Sci. USA 114, 2131–2136 (2017) Paper

Adaptable materials Memory, learning and computation in synthetic systems

We ask the same questions of synthetic systems. How can physical systems store, retrieve and process information through their material properties? Using ideas from Hopfield networks and associative memory, we study how non-equilibrium dynamics can enhance memory in elastic and disordered systems. We also ask whether chemical reaction networks can compute, and whether quantum coherence is a resource for computation in molecular networks. This builds on our work on self-assembly and soft matter far from equilibrium.

Research questions

  • Chemical computation. Can a chemical reaction network infer a rule from examples given at query time, without attention? (Carlos Floyd and Hector Lopez Rios, with Aaron Dinner)
  • Excitonic networks. Is quantum coherence a resource for computation in a driven, dissipative molecular network? (Matthew Du, with Carlos Floyd and Dipti Jasrasaria)

Also working on this: Billie Meadowcroft and Lissa Rennert

Papers

  1. Coherent advantage in the computational expressivity of excitonic networks
    M. Du, C. Floyd, D. Jasrasaria, S. Vaikuntanathan
    arXiv (2026) arXiv
  2. In-context learning emerges in chemical reaction networks without attention
    C. Floyd†, H. M. Lopez Rios†, A. R. Dinner, S. Vaikuntanathan
    arXiv (2026) · under revision, Nature Communications · †equal contribution arXiv
  3. Diffusion models learn underlying trends in actomyosin networks and predict behavior at unseen filament turnover
    E. Rennert, A. K. Behera, Y. Qiu, S. Vaikuntanathan
    bioRxiv (2026) · submitted to Biophysical Journal bioRxiv
  4. Learning via mechanosensitivity and activity in cytoskeletal networks
    D. S. Banerjee, M. J. Falk, M. L. Gardel, A. M. Walczak, T. Mora, S. Vaikuntanathan
    PRX Life 4, 013001 (2026) Paper arXiv
  5. Physical considerations in memory and information storage
    M. Du, A. K. Behera, S. Vaikuntanathan
    Annual Review of Physical Chemistry 76 (2025)
  6. Learning dynamical behaviors in physical systems
    R. Mandal, R. Huang, M. Fruchart, P. G. Moerman, S. Vaikuntanathan, A. Murugan, V. Vitelli
    arXiv (2024) · under revision, Nature Physics arXiv
  7. Active oscillatory associative memory
    M. Du, A. K. Behera, S. Vaikuntanathan
    J. Chem. Phys. 160, 055103 (2024) Paper
  8. Enhanced associative memory, classification, and learning with active dynamics
    A. K. Behera, M. Rao, S. Sastry, S. Vaikuntanathan
    Physical Review X 13, 041043 (2023) Paper
Driven, dissipative excitonic networks as a computational platform (Du et al., arXiv 2026).
Driven, dissipative excitonic networks as a computational platform (Du et al., arXiv 2026).

Generative AI + Stat mech Connections between modern AI and non-equilibrium physics

We explore connections between modern AI architectures and non-equilibrium physics. Score-based generative diffusion can be enhanced by active noise sources, chemical reaction networks can perform in-context learning, and ideas from statistical mechanics illuminate the design of machine learning algorithms.

Research question

  • Active diffusion. What changes when the noise that destroys the data is persistent, as in active matter, rather than white? (Akashnathan Aranganathan and Disha Sanwal)

Papers

  1. Non-equilibrium active noise enhances generative memory in diffusion models
    A. K. Behera, A. Lamtyugina, A. Nandy, D. Goto, D. Sanwal, C. Floyd, S. Vaikuntanathan
    Nature Communications, in press (2026) arXiv Research Square
  2. Diffusion models learn underlying trends in actomyosin networks and predict behavior at unseen filament turnover
    E. Rennert, A. K. Behera, Y. Qiu, S. Vaikuntanathan
    bioRxiv (2026) · submitted to Biophysical Journal bioRxiv
  3. In-context learning emerges in chemical reaction networks without attention
    C. Floyd†, H. M. Lopez Rios†, A. R. Dinner, S. Vaikuntanathan
    arXiv (2026) · under revision, Nature Communications · †equal contribution arXiv
  4. Connection between Hebbian unlearning and steady states generated by nonequilibrium dynamics
    A. K. Behera, M. Du, U. K. Jagadisan, S. Sastry, M. Rao, S. Vaikuntanathan
    PRX Life (2025) Paper arXiv
  5. Sequential pattern retrieval: New representations inspired by non-equilibrium physics and associative memory models
    M. Du, A. K. Behera, M. Rao, S. Vaikuntanathan
    ICLR 2025 Workshop on New Frontiers in Associative Memories (NFAM) (2025) OpenReview
  6. Enhanced associative memory, classification, and learning with active dynamics
    A. K. Behera, M. Rao, S. Sastry, S. Vaikuntanathan
    Physical Review X 13, 041043 (2023) Paper
Target, passive-noise and active-noise samples of a Gaussian mixture at three step sizes (Behera et al., Nature Communications, in press).
Target, passive-noise and active-noise samples of a Gaussian mixture at three step sizes (Behera et al., Nature Communications, in press).
Interested in our work? We are always looking for curious minds to join the group. Meet the group · Contact Suri