Compiling molecular ultrastructure into neural dynamics

A consortium of 25 neuroscientists and computational researchers, led by Konrad P. Kording, has published a forward-looking paper proposing a novel AI-driven framework called an 'ultrastructure-to-dynamics compiler.' The core idea addresses a major bottleneck in modern neuroscience: while technologies like electron microscopy and molecular imaging can now map brain structure at synaptic resolution—capturing not just where synapses are but what molecules they contain—this wealth of anatomical data has remained largely disconnected from understanding how those circuits actually function. The cost of such mapping is falling exponentially, creating a data deluge without a clear path to dynamical insight.

The proposed compiler is conceptualized as a learned mapping, likely built with machine learning models, that would ingest 'molecularly annotated ultrastructure'—the detailed, labeled anatomical maps—and output the key physiological parameters needed to run biophysical simulations. These parameters include synaptic efficacies (strengths) and local ion channel conductances, which dictate how neurons and circuits respond to stimuli. The critical requirement is paired training datasets where the same neural tissue is both imaged at high resolution and subjected to physiological recordings that measure its dynamic responses to perturbations.

If successfully developed, this compiler would fundamentally shift the neuroscience workflow. Instead of manually inferring function from structure or building models based on generic averages, researchers could automatically generate personalized, uncertainty-aware models of specific neural circuits directly from their anatomical blueprints. This transforms the structure-to-function quest from a descriptive, observational program into a predictive, engineering-oriented one. The ultimate promise is the ability to simulate how a specific, real neural circuit computes information and, crucially, to forecast the effects of experimental interventions, like a drug or a stimulation protocol, on its dynamics before ever touching the living tissue.