Researchers have developed a dimensionality-reduction technique called Rotational Component Analysis (RCA) that identifies low-variance rotational modes in brain-wide activity, even from individual trials. They applied it to motor-cortical spike trains from monkeys and humans, cortex-wide calcium imaging in mice, and human fMRI and MEG data, finding that spontaneous rotational dynamics appear consistently across species and recording methods.
In mice, a rotational mode observed during rest predicted cue-evoked behavior: the phase of the rotation when a randomly timed cue arrived determined whether the animal subsequently pulled a lever. The authors reproduced this effect using a recurrent neural network and a minimal two-dimensional model, suggesting that the intrinsic rotational mode acts as a dynamical constraint on response and learning. They also identified double-rotation modes with two centers of rotation that separate preparatory from task-related dynamics in both mouse and human data.
The same kinds of rotational modes organized neural activity during a non-motor human task, and a measure of rotational irreversibility captured early psychosis even when principal component analysis showed no group difference. The authors position RCA as a general tool for detecting hidden nonreciprocal dynamics in biological and physical datasets.