Zap and Zip. The Perturbational Complexity Index on Open Mouse EEG

Marcello Massimini’s group asked a simple question. If you knock on the brain, how complex is the echo? A conscious brain answers a single pulse with a response that spreads to many areas and keeps changing over time. An unconscious one answers with a short, simple wave, or with one large wave that every area shares. Turned into a number, this is the perturbational complexity index, PCI, informally called zap and zip. In a benchmark of 150 people, conscious and unconscious, its authors found a threshold that sorted every one of them correctly, and it has since been used to assess patients who cannot move or speak.

This console shows that measure on released data from mice. Each mouse received single electrical pulses into the secondary motor cortex, first awake and then under the anaesthetic isoflurane, while 30 electrodes on the skull recorded the EEG. The responses, the complexity values and every line on the canvas come from the released recordings.

What the recording shows

Awake, the pulse starts a response that spreads over the electrodes and keeps changing. Under isoflurane the same pulse into the same place gives a short, simpler response that fades. Leslie Claar, Christof Koch and colleagues at the Allen Institute recorded these mice with EEG and, in most of them, with Neuropixels probes in the cortex and the thalamus at the same time. When they put the pulse into the deep layers of the cortex, the awake response had two peaks, near 25 and near 180 milliseconds. From the timing of cortical and thalamic spikes they inferred that the late peak needs bursts of thalamic firing that drive the cortex again. Under anaesthesia that late peak is missing. Five of the six mice in this console had the pulse in the superficial layers, which the paper found gives smaller and simpler responses than the deep layers.

How PCIst counts

The console uses PCIst, the state transition version of PCI by Renzo Comolatti and colleagues (2019). It works on the averaged response of every electrode and has two steps.

• First it finds the few patterns across the electrodes that carry most of the response, the principal components, and keeps those that stand clearly above the activity before the pulse.
• Then, for each pattern, it counts how often the signal moves from one level to another, at every scale from small to large. It counts the same in the second before the pulse, which has no response in it, and keeps the scale where the response has the most transitions over and above that baseline. PCIst is the sum of these counts over the kept patterns.

A response that spreads to many patterns and keeps changing scores high. A response that is one simple wave scores low, however large it is. Chapter 2 shows each step on a real mouse.

What the console shows and does not show

• Six mice of the 24 in the release. These are the six sessions recorded with EEG alone, which are small enough to process. The others, of 12 to 27 GB each, carry Neuropixels data as well. In five of the six the pulse went into the superficial layers of the secondary motor cortex, and in one into the deep layers. Each mouse was recorded at its own currents, between 20 and 100 uA, and chapter 3 compares every mouse at its strongest pulse.
• The values are this site’s. PCIst here is our own implementation, written from the published method and tested against values derived by hand. It uses the parameters the paper states. The baseline and the response are 0.8 seconds each, the minimum signal to noise ratio is 1.6 and k is 1.2. The trial count is the same for both states. The paper does not say which current its PCIst values use, so the console offers every current a mouse was recorded at. For comparison, Claar and colleagues report 35.6 +/- 4.8 awake for superficial stimulation in 13 mice, and 16.5 +/- 2.4 under anaesthesia over all 31 sessions.
• Electrodes. The paper removed noisy electrodes by eye. The console uses the release’s own flag of valid electrodes, which may not match that choice exactly.
• A mouse is not a person. PCI was calibrated in people, and its threshold for consciousness does not carry over to mice. What carries over is the direction. In every mouse of the paper, PCIst was higher awake than under anaesthesia.

Why this belongs on this site

This project looks for consciousness as an emergent property of physical systems, which means it needs measures that do not depend on a person saying what they experience. PCI is one of very few such measures with clinical evidence behind it. The project’s own code carries a perturbational complexity instrument for its agent, the earlier Lempel-Ziv version of PCI by Adenauer Casali and colleagues, in perturbational_complexity.py. On the trained agent it has so far found that the stimulated stage barely responds, and the project’s results record calls the instrument unproven (pci_trained_2026_08.md). This console shows what the measure does on a real brain, which is the reference any reading on the agent has to be held against.

Sources

• Casarotto, S., Comanducci, A., Rosanova, M. and colleagues (2016). Stratification of unresponsive patients by an independently validated index of brain complexity. Annals of Neurology 80(5), 718 to 729. doi:10.1002/ana.24779
• Claar, L. D., Rembado, I., Kuyat, J. R., Russo, S., Marks, L. C., Olsen, S. R. and Koch, C. (2023). Cortico-thalamo-cortical interactions modulate electrically evoked EEG responses in mice. eLife 12, RP84630. doi:10.7554/eLife.84630
• Data: DANDI Archive, dandiset 000458, licence CC BY 4.0.
• Comolatti, R., Pigorini, A., Casarotto, S. and colleagues (2019). A fast and general method to empirically estimate the complexity of brain responses to transcranial and intracranial stimulations. Brain Stimulation 12(5), 1280 to 1289. doi:10.1016/j.brs.2019.05.013
• Casali, A. G., Gosseries, O., Rosanova, M. and colleagues (2013). A theoretically based index of consciousness independent of sensory processing and behavior. Science Translational Medicine 5(198), 198ra105. doi:10.1126/scitranslmed.3006294