Uncovering putative neural mechanisms of neurotherapeutic impacts on EEG using the Human Neocortical Neurosolver www.biorxiv.org/content/10.64898/2026.04...

Special thanks to the team for helping put this together!✨✨

Stephanie Jones @jonescompneurolab.bsky.social , David Zhou @dvwz.bsky.social, Austin Soplata @asoplata.bsky.social, Katharina Duecker @katduecker.bsky.social, Carolina Fernandez Pujol, Joyce Gao, Dylan Daniels @dylansdaniels.bsky.social

This work represents a new research direction the team is taking. If you study EEG biomarkers of neurotherapeutics we'd love to start a conversation!

Insights gained from biophysical modeling with the @hnnsolver.bsky.social software can help at many decision points in the neurotherapeutic development pipeline, including building evidence of target engagement, and for target selection and dosing

The default HNN model is used as a starting point to test hypotheses by either manually altering the values of the chosen model parameters, or using automated optimization and inference algorithms. Differences in parameter values pre-to post-treatment correspond to model-based predictions.

In this new preprint from the @hnnsolver.bsky.social team, we lay out a protocol for how biophysical modeling can link treatment-induced changes in EEG to cell- and circuit-level mechanisms of action

Biophysical modeling allows for testing mechanistic hypotheses that explain how EEG biomarkers emerge and change with drugs. Hypotheses about which drug mechanisms lead to distinct brain activity patterns must be constructed, and corresponding model parameters identified.

Measuring pre- to post/treatment EEG biomarkers has the potential to help with this goal, but current methods fail to link these biomarkers to concrete mechanisms of actions that causally alter the EEG. This missing link hinders the utility of EEG in advancing effective neurotherapeutics

Modeling EEG biomarkers begins with picking a specific brain signal that is reliably different between patient populations. An example of a hypothetical EEG biomarker is an auditory event related potential (ERP) that is suppressed in post-treatment (red) relative to pre-treatment (blue).

One of the major challenges in neurotherapeutic development is understanding how treatments are impacting individual brain networks

Biophysical modeling to develop and test mechanistic hypotheses underlying pharmacological EEG biomarkers. Top: Modeling EEG biomarkers begins with picking a specific brain signal that is reliably different between patient populations. An example of a hypothetical EEG biomarker is an auditory event related potential (ERP) that is suppressed in post-treatment (red) relative to pre-treatment (blue). Middle: Biophysical modeling allows for testing mechanistic hypotheses that explain how EEG biomarkers emerge and change with drugs. Hypotheses about which drug mechanisms lead to distinct brain activity patterns must be constructed, and corresponding model parameters identified. Bottom: The default HNN model is used as a starting point to test hypotheses by either manually altering the values of the chosen model parameters, or using automated optimization and inference algorithms. Differences in parameter values pre-to post-treatment correspond to model-based predictions.

Happy to share a new preprint from the @hnnsolver.bsky.social team!🧠💻🎉

"Uncovering putative neural mechanisms of neurotherapeutic impacts on EEG using the Human Neocortical Neurosolver"

📝 www.biorxiv.org/content/10.6...

New preprint! 🧠
How do RNNs learn abstract rules from sequences, independent of specific stimuli?

By Vezha Boboeva, with Alberto Pezzotta & George Dimitriadis

"From sequences to schemas: low-rank recurrent dynamics underlie abstract relational representations"
www.biorxiv.org/content/10.6...

Thrilled to see the second paper of the lab out 🤩 Check it out if you need a method to infer causality from neural data, even when the signal is short!

Paper: joss.theoj.org/papers/10.21...
Code: github.com/CMC-lab/Tran...

Want a dataset to test ideas on neural basis of decision making or how areas interact as we make choices? Check out our data published today @rudebecklab.bsky.social. >16,000 single neurons from 22 anatomically confirmed areas in macaques performing a decision task. www.nature.com/articles/s41...

Constructing connectome-based neural networks

Really excited to share a new preprint!

𝗛𝗼𝘄 𝗱𝗼 𝗯𝗿𝗮𝗶𝗻𝘀 𝘀𝘁𝗮𝘆 𝗿𝗼𝗯𝘂𝘀𝘁 & 𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝘄𝗵𝗶𝗹𝗲 𝗯𝗲𝗶𝗻𝗴 𝗶𝗻𝗰𝗿𝗲𝗱𝗶𝗯𝗹𝘆 𝘀𝗽𝗮𝗿𝘀𝗲? We explore how!
www.biorxiv.org/content/10.6...

We built Connectome-based Neural Networks (CoNNs) using Drosophila wiring (larva&adult) & compared with random networks with same sparsity.

Preprint out arguing that we should build the techology to translate (compile) molecularly annotated connectomes into dynamics. I think this is incredibly important. arxiv.org/abs/2603.25713

This work started when Christian Pehle, @tonyzador.bsky.social, and I found that training sequentially (consuming/producing one spike at a time) in JAX was prohibitively slow. You certainly can consume multiple spikes in parallel—TLDR; just use an associative scan—but... 2/N

🧵 New preprint led by @bingbrunton.bsky.social, @elliottabe.bsky.social, @lawrencehu.bsky.social

We gave a worm brain control of a fly body and it walked

What did we learn? Nothing, other than deep reinforcement learning is effective

We call it the digital sphinx

www.biorxiv.org/content/10.6...

Our paper “Multifidelity Simulation-based Inference for Computationally Expensive Simulators” has been accepted at ICLR 2026! 🥳

We hope this can be a practical solution for anyone analysing and doing inference on computationally expensive simulators.

Paper: openreview.net/pdf?id=bj0dc...

Image showing the paradigm for recording single neurons in a variety of brain regions in humans, while single-pulse TMS was applied to dlPFC. One subpanel shows that single neurons were able to be resolved very early (8ms) after the single-pulse stimulation. Another subpanel shows two example neuron waveforms, along with their inter-spike interval distributions. The final subpanels show spike density functions overlaid on raster plots for the same neuron, separately for active and sham stimulation, in order to demonstrate that active (but not sham) stimulation increased activity in this example neuron.

We have a new preprint examining single neuron responses in humans undergoing TMS (www.biorxiv.org/content/10.6...). The excellent Charlie Dickey led the way, and I had the privilege to co-supervise this work with @coreykeller.bsky.social and @aaronboes.bsky.social 🧵 (1/7) 🧠🟦 🧠💻

A Spatially Structured Spiking Network Model of Beta Traveling Waves and Their Attenuation in Motor Cortex www.biorxiv.org/content/10.64898/2026.03...

Representation Biases: Variance Is Not Always a Good Proxy for Importance A central approach in neuroscience is to analyze neural representations as a means to understand a system's function, through the use of methods like principal component analysis, regression, and repr...

Pleased to share that our paper "Representation Biases: Variance is Not Always a Good Proxy for Importance" is now out as Theory/New Concepts paper in eNeuro!
www.eneuro.org/content/13/3... 1/

In the future we hope to apply this framework to biophysical models of EEG generation like @hnnsolver.bsky.social providing a link between biomarkers and behaviorally relevant neural dynamics

Special thanks to my amazing advisor Stephanie Jones @jonescompneurolab.bsky.social for all your support!

Figure showing neural trajectories to modified versions of the NMDA receptor arriving at the dendrite where either 1) the receptor just exhibits slow time constants, or 2) the receptor just exhibits a magnesium block. Neither variant was successfully trained to solve the task, suggesting that the unique properties of NMDA receptors are necessary for the emergence of fixed point attractors.

We also went deeper into what aspects of the NMDA receptor promote the emergence of attractor dynamics and found that the unique combination of slow receptor time constants and a magnesium block are both critical for task performance

Figure of loss curves during training for different network variants where extrinsic inputs were fixed. Limiting extrinsic inputs to signal through NMDA receptors produced the best trained networks. AMPA inputs at the dendrite failed to solve the task.

In line with previous literature, slow time constant synapses (i.e. NMDA receptors) were shown to be extremely important for the emergence of attractor dynamics

What’s surprising is that this result holds even when the network’s connectivity and active ion channels are freely changed

Schematic figure illustrating how different biophysical properties are evaluated for their importance to generate fixed point attractors, 1) fix extrinsic inputs (i.e. NMDA or AMPA inputs to the dendrite or soma), 2) train cell and network parameter (i.e. all other parameters of the model are freely modified), 3) evaluate task performance (the emergence of distinct fixed point attractors for each input is considered a successfully trained network)

We do this by selectively constraining certain biophysical properties, and allowing all others to be freely optimized

Visualization of the different scales of the biophysical model used in the paper. 1) Subcellular where voltage traces of a dendritic action potential are shown alongside the morphology of a multicompartment neuron, 2) cell-level where a spike raster shows how a single neuron exhibits preferential spiking to different inputs to the network, 3) network-level where a spike raster and neural trajectory plot both show that the network produces sustained changes in activity to inputs.

Using the new modeling framework Jaxley, we attempt to characterize how fixed point attractors emerge in RNNs composed of biophysically detailed neurons

Currently there’s a lot of work understanding attractor dynamics in RNNs, as they are readily linked to cognitive tasks and behaviors

Since RNNs typically contain simplified neurons, it is unclear how biophysical properties of realistic neurons impact the emergence of attractor dynamics

Multipanel figure illustrating the key components of the paper: 1) biophysical reservoir computing where a network of biophysically detailed excitatory and inhibitory neurons are randomly connected, receive a brief input, and produce a sustained spiking pattern in response, 2) an illustration of the task the biophysical reservoir computer is trained on: a simplified working memory task where the network must produce distinct fixed point attractors in response to different inputs.

Happy to share a new preprint from my PhD thesis! “A novel framework for expanding RNNs with biophysical detail to solve cognitive tasks” 🧠💻

📝 www.biorxiv.org/content/10.6...

@jonescompneurolab.bsky.social

Deep-learning-assisted simulation of a cortical circuit: integrating anatomy, physiology and function www.biorxiv.org/content/10.64898/2026.03...

We've been saying for a while now that beta bursts might be functionally heterogeneous. Check out @quentinmoreau.bsky.social's paper where we show that different types of bursts have different relationships to behavior in a sensorimotor adaptation task 👇👇👇

Mission: record hippocampal place cells in zero gravity
Crew: 3 rats with electrode arrays
Vehicle: Space Shuttle Columbia
Status: data now publicly available on DANDI, 28 years later

The ratstronauts' mission is finally complete. 🐀🚀 h/t NASA

about.dandiarchive.org/blog/2026/02...