Real-time neuroscience: closing the loop between data and experiment
In many neuroscience experiments, data are collected first and analyzed later. Neural activity is recorded, behavior is tracked, and only after the experiment ends do we learn which neurons were important, which stimuli were informative, or which perturbations would have revealed something new. By then, the experiment is over—and the opportunity to adapt is gone.
Anne Draelos and coauthors introduce "improv", a software platform designed to make experiments adaptive. Instead of separating data collection from analysis, improv allows the experiment to respond to the data as they arrive. Imaging, behavioral tracking, modeling, and stimulation control all share a live memory space, so models can be updated continuously and used to guide the experiment in real time.
This means the experiment can ask smarter questions as it unfolds. While recording from the zebrafish brain, improv can estimate which neurons respond to motion and immediately target them with optogenetic stimulation. While observing spontaneous behavior and neural activity, it can identify latent variables linking the two and adjust the experiment to probe them further. During electrophysiology in motor cortex, it can learn the evolving neural trajectory and predict where it is heading, opening the door to precisely timed interventions.
The core idea is simple: analysis becomes part of the experiment, not something that happens after it. By closing the loop, improv turns experiments into dynamic conversations with the brain, where hypotheses can be updated continuously and causal tests can be performed when they are most informative.
This points toward a new generation of neuroscience experiments—faster, more efficient, and more interactive—where the limiting factor is no longer how much data can be recorded, but how intelligently it is used in real time.
Paper:
nature.com/articles/s41467-0…