Neural Timescale Analysis · good_isttc

Intrinsic timescales
across brain regions

Spontaneous spiking activity from 154 sessions. Autocorrelation decay fitted with 1–4 exponential components. All units passed quality filter (r² > 0.5, CI excludes zero, ACF decline 50–200 ms).

23,660
Total units
4.3 Hz
Median firing rate
300 ms
Median τ effective
199
Unique brain regions
01
Timescale × Brain Region
Median effective timescale (τ) for the 15 most-sampled regions, sorted descending.
τ > 500 ms 200–500 ms τ < 200 ms
02
Timescale × Major Brain Division
Comparing median effective τ across grouped forebrain, midbrain, and hindbrain structures (Top 15 regions mapped).
Forebrain (Ctx/Hippo/BG/Thal) Midbrain (MRN/PRNr) Hindbrain (SPVI)
Hierarchy validated: Aligning with Shi et al. (2025), there is a ~8× jump in timescale magnitude from the forebrain (110 ms median) to the midbrain complex (868 ms), establishing a massive hierarchical gradient in temporal integration.
03
Timescale × Firing Rate
Median τ effective per firing rate bin, with unit count overlay.
Median τ (left axis) Unit count (right axis)
04
Timescale × Local Variance (LV)
LV measures inter-spike interval regularity. LV ≈ 1 is Poisson-like; LV < 1 is more regular; LV > 1 is more bursty.
Median τ effective (ms)
05 — 06
τ Distribution & N Timescales
Left: histogram of τ effective across all 23,660 units. Right: number of exponential components needed to fit the ACF.
Unit count per τ bin
n_timescales distribution
07
Timescale Components — τ₁ through τ₄
Median value of each fitted timescale component across all neurons.
τ₁ — 1st comp
52.4 ms
n = 23,660
τ₂ — 2nd comp
1,034 ms
n = 17,203
τ₃ — 3rd comp
2,929 ms
n = 3,151
τ₄ — 4th comp
4,489 ms
n = 22
Median τ₁ and τ₂ per brain region
Median τ₁ Median τ₂
08
Multiscale Architecture: τ₁ vs τ₂ Relationship
Testing the hypothesis that fast components (τ₁) drive regional differences while slow components (τ₂) act as universal, state-dependent baseline dynamics.
Top 15 Regions (Log-Log Scale)
Regional vs Universal: While τ₁ varies widely across regions (from ~19ms in CA1 to ~278ms in V), τ₂ maintains a proportionally higher baseline in regions with long τ₁, demonstrating a coupled multiscale architecture where local circuit properties (τ₁) scale alongside slower network states (τ₂).
09
Tau₂ Power-Law Distribution
Log-log histogram of τ₂ across the population to test the theoretical claim of scale-free dynamics (exponent ≈ -2).
τ₂ Unit Count (Log-Log Scale)
Deviation from Scale-free Dynamics: Unlike the theoretical claim of a pure power-law (where the line should descend sharply and linearly), the empirical data shows a broad plateau between 1,000 and 3,500 ms. This suggests that slow fluctuations do not follow strict criticality, but rather cluster within a preferred "slow state" frequency band across the brain.
10
Fit Reliability & Amplitude Metrics
Evaluating model confidence (R² per region, CI width) and coefficient magnitude (c₁ vs τ₁).
Median R² per Region (Top 15)
R² > 0.85 R² < 0.85
Amplitude (c₁) vs τ₁ (Top 30)
Median Region Data
Confidence Interval Width (95%) vs Effective τ
Estimation limits & Amplitude tradeoff: Uzun timescale tahminleri doğası gereği daha geniş güven aralıklarına (CI) sahiptir. Daha da önemlisi, c₁ ve τ₁ arasında güçlü bir ters orantı vardır. Kısa τ₁'e sahip bölgeler (örn. CA1, SSp) otokorelasyonda çok daha yüksek bir başlangıç genliğine (c₁ ≈ 0.20) sahipken, beyinsapı gibi uzun τ₁'li bölgelerde bu genlik zayıflar (c₁ ≈ 0.07).