Hypothesis
Spectral coherence quantifies phase locking between oscillating neuronal groups and so tracks synchronization-mediated information flow between networks.
What failed
Using simulated coupled phase oscillators and interacting gamma-generating spiking networks, coherence deviated substantially from the expected phase locking over a large parameter range and did not converge on the expected value as signal-to-noise ratio rose; it failed particularly in the partially synchronized regime that the authors expect to be most common in the brain, and often failed to reflect changes in information flow measured by transfer entropy.
Broken assumption
Coherence in a frequency band measures phase locking between neural oscillators.
Why
The authors attribute the failure to the fast frequency and amplitude changes that synchronization forces themselves induce, and propose a phase-locking value based on Hilbert transform after singular spectrum decomposition instead.
From the published paper: Lowet et al. PLoS ONE, 2016. ↗CC BY