Purushottam Dixit

🧵 Our new paper is out in Physical Review Research! Can destroying a signal make it easier to sense Surprisingly, yes! We show that ligand endocytosis can reshape extracellular chemical gradients and enhance directional information.

View full thread

6/This suggests a new function for receptor endocytosis. A receptor is not simply reading an extracellular signal. By removing ligand, receptors can actively reshape the signal outside the cell before the cell reads it.

7/More broadly, cells may process information not only through intracellular signaling networks, but by modifying their extracellular environment. Sometimes, destroying a signal is part of the computation. Paper: journals.aps.org/prresearch/a...

4/The mechanism turns out to be particularly simple. As receptor-mediated endocytosis increases, the overall ligand concentration decreases. Even the absolute difference in concentration between the front and back of the cell decreases. And yet…

5/…the relative difference between the front and back increases. Endocytosis makes the extracellular signal weaker but increases its directional contrast. Signal destruction can actually make direction easier to read.

2/ The ingredients are simple: One cell secretes a chemical signal -> The signal diffuses -> A nearby cell detects it through receptors. But receptors often remove ligand from the environment through endocytosis.

3/There is a beautiful precedent for this idea. Work from Andrew Mugler and colleagues showed that signal degradation in 3D space can improve spatial sensing by sharpening otherwise diffuse signals. That made us wonder: could surface-based degradation?

1/ Cells often migrate toward one another even when there is no externally imposed chemical gradient. This raises a basic question: How does a cell know where another cell is?

journals.aps.orgDiscovering interpretable low-dimensional dynamics using maximum entropyModels (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-dimens...

21 Sept

Our paper on low dimensional max ent dynamics is out in Phys Rev E. We introduce Edwin. It compresses high-dimensional dynamics into a few latent variables and discovers symbolic equations for those variables: journals.aps.org/pre/abstract...

Tested on 2D Brownian motion, Ornstein-Uhlenbeck, Smoluchowski self-assembly, spiking neural populations, low-rank RNNs, and one noisy experiment: aggregating RNA-liposome complexes, where it recovers a power-law exponent that maps onto known aggregation physics.

21 Sept

Led by @michaelcchung.bsky.social, with Tarran Mohan, and with Juan Guan, whose lab drove this.

Edwin (after ET Jaynes) compresses data using the dynamic maximum entropy principle by performing dimensionality reduction using Max Ent. This splits the data into time-dependent latents Z(t) and feature-dependent latents Y.

21 Sept

It then learns sparse symbolic models for both latents and features; ODEs for latents and interpretability for feature.

PCA and autoencoders optimize reconstruction, and the latents they return rarely correspond to anything measurable. Equation discovery of the SINDy family assumes you already know the state variables. In high-dimensional data you usually do not.

Purushottam Dixit reposted

APS journals

@apsjournals.aps.org

17 Sept

New in PRX Life: microbial ecosystems are low-dimensional. After studying about 200 gut microbiome datasets, the authors found that niche dimensionality was below predicted value, upending previous models. Read the paper: https://go.aps.org/46qJqBa

journals.aps.orgLow-Dimensional Coexistence in Complex Microbial EcosystemsAnalysis of nearly 200 human gut microbiome datasets shows that microbial coexistence is governed by a few effective ecological dimensions, rather than the high-dimensional niches assumed by prevailing theory.

1/ Our new paper is out in PRX Life. Physicists' often treat microbiomes as disordered systems with coexistence in very high-dimensional niche spaces. We asked whether empirical data supports this picture, and found the data point somewhere else: journals.aps.org/prxlife/abst...

6/ Why? Low dimensionality tracks metabolic competition and lower diversity, exactly mirroring the "niche dimensionality hypothesis". Our simulations recover it only with correlated resource supplies plus metabolic tradeoffs. Dimensionality is lower in simplified diets and in disease.

7/ Encouragingly, others (e.g. @asanchezlab.bsky.social) are also finding similar signatures of non-random interactions in microbiomes!

4/ Our model showed that the inferred niche axes are biologically interpretable. Species loadings predict metabolic capabilities from genome-scale models, embeddings predict measured metabolites.

5/ Notably, across ~200 human studies (> 100000 individual ecosystems), effective dimensionality was very low, around 4. Diversity-matched disordered (spin-glass-like) models predict a much larger number (~15). Real communities are far lower-dimensional than the random-interaction picture assumes.

2/ The spin-glass view models microbiomes as many species with random, weak, and all-to-all interactions. It reproduces broad patterns like species abundance distributions. Yet, this high-dimensionality assumption is rarely tested directly.

3/ To test the dimensionality empirically, we went back to a classical ecological idea: Hutchinson's niche dimensionality. We made it measurable, inferring an effective niche dimensionality straight from abundance data with a low-rank model that can be derived from consumer-resource dynamics.

cell.comAn information-theoretic framework for transcriptional regulation reveals declining communication fidelity in agingEmison et al. develop an information-theoretic framework that decomposes communication fidelity in gene regulatory networks. Applying this framework to single-cell RNA-seq data across aging tissues, they find that communication fidelity declines, driven primarily by shifting input distributions rather than increased channel noise.

1/ New paper out! Aging has been described as a loss of "information", but nobody could quantify it. We built a stat phys model for it. We found taht communication fidelity in gene regulation drops with age across multiple tissues, making it a hallmark of aging! 🧵: www.cell.com/cell-reports...

7/ The model suggests that switching a few TFs back on in silico (Ebf2, Ebf1, Aebp1, Elf3, Foxq1) restores much of the youthful information transfer AND pushes gene expression back toward young, echoing how just a few factors can reset cell identity.

github.comGitHub - emisbrooke/InFlow: Information theoretic framework for quantifying trancriptional regulationInformation theoretic framework for quantifying trancriptional regulation - emisbrooke/InFlow

9/ Led by Brooke Emison, with @chriswlynn.bsky.social , Andrew Mugler & Fabrisia Ambrosio. Code is here: github.com/emisbrooke/InFlow

5/ Aging keeps the TF→target wiring largely intact but desynchronizes the joint TF activity that wiring was tuned to read. "Input mismatch," not "channel corruption." The regulators fall out of sync rather than getting miswired.

6/ Consistent with this, the TF–TF network reorganizes with age: more centralized, hub-dominated, depleted of stabilizing feedback motifs, and more fragile when you perturb it.

3/ Across 10 mouse tissues, information transfer from TFs to their targets drops with age in every single tissue. And the faster a tissue turns over its cells, the bigger the drop.

4/ The useful part: this information depends on two factors, the channel (TF→gene wiring) and the input distribution (how TF activity is distributed). Because of the analytically tractable model, we could ask which one breaks with age. The answer surprised us: it's not the wiring.

2/ We treat the gene regulatory network as a communication channel between transcription factors (TFs) and their target genes (TGs) and compute the mutual information between TFs and TGs from single-cell data. A conditional maximum-entropy model makes this tractable in high dimensions.

cell.comLow-affinity ligands of the epidermal growth factor receptor are long-range signal transmitters in collective cell migration of epithelial cellsDeguchi et al. find that low-affinity EGFR ligands propagate faster and farther than high-affinity ligands in epithelial cells. They demonstrate that EREG, a low-affinity ligand, contributes to skin w...

24 July

These were the experiments we were trying to explain: www.cell.com/cell-reports...

Purushottam Dixit@pdixit.bsky.social

EGFR signaling breaks this tradeoff: low affinity ligands diffuse farther from the source (expected) but also lead to higher activity (unexpected).We show that kinetic proofreading combined with receptor/ligand degradation explains this phenomena. Work done in collaboration with an undergraduate!

International Initiative for Theoretical Ecology

24 July

Traditional models of paracrine gradients are constrained by a tradeoff driven by ligand-receptor affinity. High activity implies shorter spatial range and vice versa. Strong binders lead to higher signaling activity but also trigger receptor and ligand degradation and therefore shorter range.

bioRxiv Biophysics@biorxiv-biophys.bsky.social

Kinetic proofreading decouples signal strength and range in paracrine gradient formation https://www.biorxiv.org/content/10.64898/2026.07.17.739274v1

I will be talking about some recent work on using niche theory to model host associated microbiomes in a couple of hours!

International Initiative for Theoretical Ecology@iite-ecotheory.bsky.social

📣Tomorrow! 📣Purushottam Dixit (Yale Engineering) will present: 'Niche dimensionality drives microbial community structure'

Purushottam Dixit reposted

@iite-ecotheory.bsky.social

30 Mar

Replying to @iite-ecotheory.bsky.social

📣Tomorrow! 📣Purushottam Dixit (Yale Engineering) will present: 'Niche dimensionality drives microbial community structure'

pnas.orgPNASProceedings of the National Academy of Sciences (PNAS), a peer reviewed journal of the National Academy of Sciences (NAS) - an authoritative source of high-impact, original research that broadly spans...

How do cells know which way to move in a chemical gradient? 🧭 New work by graduate student Andrew Goetz proposes that receptors can compute direction. This new mechanism for directional sensing was published in PNAS late last year: www.pnas.org/doi/10.1073/...

The stochastic analysis is especially striking: Noise reveals an optimal basal activity set point that maximizes signal-to-noise. Too little activity → noisy. Too much → less directional contrast. Biology tunes itself in between.

Big picture: Directional sensing can be a receptor-level computation driven by diffusion + degradation + feedback. This explains puzzling experiments (e.g., why blocking endocytosis impairs chemotaxis) and suggests a general biophysical strategy across receptors families.

Add diffusion, and something remarkable happens: Receptors move from low-ligand regions to high-ligand regions, but degradation depletes them where ligand is high. This mismatch creates polarized receptor activity-directional sensing emerges naturally.

Predictions: ✔ There is an optimal diffusion rate for maximal polarization ✔ Receptors encode relative, not absolute, ligand gradients ✔ Cooperative receptor interactions can amplify weak gradients All without requiring separate “local” and “global” signaling species.

The key ingredients: • Receptor diffusion along the membrane • Basal (ligand-independent) activity • Selective degradation of active receptors Together, these create an integral feedback loop (as opposed to an IFFL in LEGI) at the cell surface.

Without diffusion, degradation implements integral feedback: receptor activity adapts perfectly to a ligand-independent set point-essential for sensing relative changes rather than absolute ligand levels.

Classic models of eukaryotic directional sensing (e.g., LEGI) assume receptors are passive: they report ligand levels, while intracellular networks compare “local vs global” signals using incoherent feedforward loops. We asked: can receptors do the computation directly?

What if aging isn’t just cellular damage, but a lossy transmission of information? We used single-cell data and physics-based models to show that as cells age, the flow of transcriptional information collapses. Here’s what that means. 👇

bioRxiv SysBio@biorxiv-sysbio.bsky.social

Decay in transcriptional information flow is a hallmark of cellular aging https://www.biorxiv.org/content/10.1101/2025.10.30.685689v1

While signs of cellular aging may be varied, across all tissues, mutual information between TFs and TGs declines with age. Aged networks show input mismatch, higher centralization, and reduced stability, patterns reminiscent of aging brains and failing ecosystems.

Given the centralization of information flow, our analysis suggests that re-tuning just a few TFs could in principle rejuvenate information flow and restore gene expression. Aging may not be just cellular damage, but a gradual communication breakdown.

bioRxiv SysBio@biorxiv-sysbio.bsky.social

Decay in transcriptional information flow is a hallmark of cellular aging https://www.biorxiv.org/content/10.1101/2025.10.30.685689v1

We analyzed single-cell RNA-seq data from multiple mouse tissues across the lifespan. This dataset captures how thousands of genes are expressed in individual cells, letting us see how regulatory communication changes with age.

We borrowed statistical physics/neuroscience models to study information transmission in noisy systems. We treated transcription factors (TFs) → target genes (TGs) as a multi-input, multi-output communication channel, and measured its fidelity using mutual information.

bioRxiv SysBio@biorxiv-sysbio.bsky.social

Decay in transcriptional information flow is a hallmark of cellular aging https://www.biorxiv.org/content/10.1101/2025.10.30.685689v1