maedoc/autotvb/scientific-validity

TVB Navigator: Scientific Validity & Parameter Regimes

Model Parameter Regimes

Validity Checks

• Does the chosen model match the scientific question? (Epileptor for seizure, JansenRit for ERP, G2D for oscillations, RWW for resting-state)
• Is connectivity speed physiologically plausible (1–10 mm/ms)?
• Is coupling strength appropriate for the model and connectivity density?
• Does the analysis match the data type? (BOLD for resting-state FC, TemporalAverage for ERP latencies)
• Are statistical comparisons using appropriate null models or surrogates?

Regime Verification

Do not accept regime claims from parameter tables alone. Require empirical verification in the notebook:

• Compute PSD and report the dominant peak frequency.
• Inspect the time-series amplitude envelope: sustained = limit cycle; decaying = stable spiral.
• If the driver claims "N Hz", the PSD peak must agree within ±1 Hz.

Focal Parameterization for Seizure Models

• Epileptor focal zone: Do not apply a uniform or randomly distributed x0 across all regions. The Epileptor requires a focal epileptogenic zone (a small subset of regions with elevated x0, e.g., -1.6) embedded in a surround of more negative x0 (e.g., -2.4 to -2.5). Flag any non-focal x0 parameterization as scientifically invalid.

Windowing & Stimulus-Locked Analysis

• Burn-in must be shorter than stimulus onset: Flag immediately if burn-in ≥ onset (e.g., 1000 ms burn-in with 500 ms onset). The evoked response must remain in the data.
• Align analysis windows with the scientific question: For stimulus-propagation studies, FC and spectral analyses should target [onset, onset + window], not arbitrary post-burn-in intervals.
• Distinguish intrinsic vs. evoked FC: ongoing oscillations produce structurally-driven FC; a post-stimulus window may show propagation. The driver must state which is being reported.