main concepts scientists sections
SimplePopularAdvanced

From Brain Scans to Brain Function express

Original: "Compiling molecular ultrastructure into neural dynamics"
· Konrad P. Kording, Anton Arkhipov, Davy Deng, Sean Escola, Seth G. N. Grant, Gal Haspel, Michał Januszewski, Narayanan Kasthuri, Nina Khera, Richie E. Kohman, Grace Lindsay, Jeantine Lunshof, Adam Marblestone, David A. Markowitz, Jordan Matelsky, Brett Mensh, Patrick Mineault, Andrew Payne, Joanne Peng, Xaq Pitkow, Philip Shiu, Gregor Schuhknecht, Sven Truckenbrodt, Joshua T. Vogelstein, Edward S. Boyden
arXiv:2603.25713 · 2026-03-26 · CC BY 4.0 · 1 min
Neurons and Cognition q-bio.QM
Scientists are learning to predict how a neuron will behave just by looking at its structure.

📄 Showing the "Simple" version — "Popular" is not ready yet. Add it to favorites to help prioritize it.

Modern microscopes see the brain in exquisite detail: not just the synapses themselves, the contact points between neurons, but also the molecules they are made of. Such images are becoming more common and cheaper to produce. But there's a catch: a pretty picture doesn't tell you how quickly a cell transmits a signal or how strongly one neuron influences another.

The authors propose building a "translator"—a program that takes a molecularly labeled image and outputs a set of numbers describing how the neuron works: synaptic strength, channel conductance. They call this translator a compiler: just as a program turns code into an executable file, this tool turns structure into dynamics.

To train such a compiler, you need paired data: an image of a brain region and a recording of how that region responds to stimulation. Once you've gathered enough pairs, you can teach the algorithm to predict physiology from anatomy alone.

Then biological simulations become more accurate: instead of hand-tuning parameters, researchers can load real connectivity maps and run the model right away. This is a step from simply describing "what is connected to what" to predicting "how this network will think."

This approach could also help in medicine: knowing how tissue structure changes in disease, you can predict how its function will change and assess the effect of treatment in advance.

It all started with pretty pictures. Now they promise to become a working tool, like a blueprint from which you assemble a functioning mechanism.

Whole graph →

The cost of mapping synapses is falling so fast that soon a detailed map of a brain region will cost no more than a routine lab test.

Concepts
neural networks in spatial mapping medical image to genomic mapping neural network molecular dynamics acceleration computation predictive model molecular dynamics simulation learning autoencoder bioinformatics mathematical model biomolecule neurotransmitter deep learning in medical imaging parameter protein design brain connectivity biological rhythm modeling biomarker data processing