PaperlayerView exact revision on arXiv

Source-linked AI summary

DyssomniaNN: Multi-Scale Spiking Neural Modeling of Human Sleep Disorders and Perceived Restedness

Daniel Szelogowski

arXiv:2610.13918v1cs.NEcs.ETcs.LGq-bio.NC

TL;DR

Objective sleep duration can diverge from subjective impairment, and prior models have not united sleep regulation across biological scales in continuous multi-day simulations. DyssomniaNN couples homeostatic and circadian drives, hypothalamic switching, and a corticothalamic spiking network. Its simulations show that restorative recovery can diverge from sleep duration, while generated spectra converge with empirical sleep recordings.

  • Problem

    Objective sleep duration and sleep efficiency may fail to track perceived impairment, and prior models had not united sleep-regulatory scales in continuous multi-day simulations of distinct dyssomnias.

    Source: §1 · pp. 1–2

  • Method

    DyssomniaNN couples Borbély’s two-process regulation, Saper’s hypothalamic flip-flop circuit, and a recurrent corticothalamic LIF network across multi-day simulations.

    Source: §1 · p. 2

  • Results

    21.15 h of sleep duration coexisted with R = 37.6% restoration in paradoxical insomnia, while synthetic spectra converged with empirical PSG spectra across canonical EEG bands.

    Source: §1 · p. 2, §8 · p. 15

  • Takeaways & Limitations

    Total sleep time is an incomplete proxy for restorative sleep when cortical hyperarousal or subcortical boundary instability is present.

    Source: §8 · p. 15

    The corticothalamic network is a lumped single-channel representation, and future extensions are proposed for spatial connectivity and more detailed REM transitions.

    Source: §7.1 · p. 14

Abstract

from arXiv · show

Clinical sleep medicine repeatedly documents profound divergences between objective sleep duration and subjective daytime impairment, notably exemplified by sleep state misperception in paradoxical insomnia and boundary instability in narcolepsy type 1. Here, we present DyssomniaNN: a multi-scale biophysical computational framework coupling Borb\'ely's two-process regulation of sleep homeostasis and circadian drive with Saper's hypothalamic reciprocal flip-flop circuit and a recurrent corticothalamic leaky integrate-and-fire spiking neural network. We formalize five clinical cohorts across an identical 72-hour photoperiod: healthy control, circadian rhythm sleep-wake disorder (delayed-phase), chronic insomnia, paradoxical insomnia, and narcolepsy type 1. The synthetic local field potential converges closely with empirical human electroencephalography spectra from the PhysioNet Sleep-EDF database across all canonical frequency bands (spectral mean absolute error of 0.33--0.35). Furthermore, we introduce the Well-Rested Index to quantify restorative efficacy. Simulations reveal that sleep duration fails to predict restorative recovery: paradoxical insomnia achieves normal total sleep time (21.15 h) yet yields low restoration (37.6%) from unsuppressed cortical beta-band intrusion, whereas hypocretin deficiency in narcolepsy produces severe nocturnal fragmentation (781 min of wake after sleep onset) and 35 sleep-onset rapid eye movement periods. DyssomniaNN provides an in silico platform for mechanistic sleep research and virtual neuropharmacology.

1 Introduction

Objective sleep duration can diverge from subjective impairment, while existing sleep models leave multiple biological scales disconnected. DyssomniaNN couples those scales across multi-day simulations and evaluates sleep restoration as well as EEG spectra.

1 Introduction

  • Paradoxical insomnia can involve subjective insomnia and daytime exhaustion despite normal recorded sleep duration, sleep onset latency, and macro-architecture; narcolepsy can involve severe sleepiness despite equal or greater 24-hour total sleep time.These examples illustrate why polysomnographic total sleep time and sleep efficiency may not track perceived impairment or neurocognitive recovery.Source: §1 · p. 1
  • Prior models separately represented circadian and homeostatic drives, hypothalamic switching, or local cortical rhythms, without uniting these scales in continuous multi-day simulations of distinct dyssomnias under one photoperiod.The paper specifies a shared 16-hour light/8-hour dark schedule.Source: §1 · p. 2
  • DyssomniaNN couples Borbély’s two-process regulation, Saper’s hypothalamic flip-flop switch, and a corticothalamic LIF network, parameterizing five clinical phenotypes across 72 hours.The phenotypes include healthy sleep, delayed-phase CRSWD, chronic insomnia, paradoxical insomnia, and narcolepsy type 1.Source: §1 · p. 2
  • Synthetic corticothalamic EEG spectra quantitatively converged with empirical PhysioNet Sleep-EDF spectra across Delta, Theta, Alpha, Sigma, and Beta bands.The section reports quantitative spectral convergence but gives no error value here.Source: §1 · p. 2
  • The Well-Rested Index (R ∈[0, 100%]) captures restorative efficacy, which the paper says is dissociated from total sleep time and governed by slow-wave preservation, cortical hyperarousal, and circadian misalignment.This index addresses the gap between objective sleep measures and perceived daytime impairment.Source: §1 · pp. 1–2

2 Related Work

Prior work modeled sleep regulation at distinct phenomenological, hypothalamic, and neuronal scales. DyssomniaNN’s motivation is to connect those scales over continuous multi-day trajectories.

2 Related Work

  • Borbély’s two-process model describes sleep timing and duration through interacting homeostatic pressure and circadian processes, but lacks direct representation of subcortical state-transition nuclei.Homeostatic pressure accumulates during wake and discharges during sleep, while circadian drive is associated with the suprachiasmatic nucleus.Source: §2 · p. 2
  • The VLPO and lateral hypothalamic area form a mutually inhibitory flip-flop circuit; reciprocal inhibition supports bistable sleep and wake states, while orexin loss is associated with rapid involuntary transitions.The passage describes inhibitory VLPO projections to arousal centers and reciprocal inhibition of the VLPO by monoamines.Source: §2 · p. 2
  • Thalamocortical loops support delta oscillations and N2 spindles, while primary insomnia is associated with elevated NREM beta and gamma activity.Existing spiking-network models typically cover seconds or minutes rather than continuous multi-day homeostatic and circadian trajectories.Source: §2 · pp. 2–3

3 Biophysical Framework

DyssomniaNN links homeostatic-circadian regulation to hypothalamic sleep-wake switching and a recurrent corticothalamic spiking network. Its synthetic LFP is derived from network activity and evaluated against empirical PSG spectra.

3 Biophysical Framework

  • Three hierarchical layers couple homeostatic-circadian regulation, hypothalamic bistable switching, and a recurrent corticothalamic spiking neural network.The subcortical arousal level gates the LIF network, whose synthetic LFP is evaluated against empirical PSG spectra.Source: §3 · p. 3, Figure 1 · p. 3

3.1 Two-Process Homeostatic and Circadian Model

  • During wakefulness, S(t) rises toward saturation with an 18.2 h time constant; during sleep, it decays toward zero with a 4.2 h clearance time constant.Circadian drive C(t) is a periodic oscillator coupled to environmental lux; its phase offset supports modeling chronotypes and phase-shift disorders.Source: §3.1 · p. 4

3.2 Reciprocal Hypothalamic Flip-Flop Dynamics

  • Mutual inhibition between VLPO and arousal populations creates a bistable switch, while depleted hypocretin shrinks the wake attractor’s basin and can destabilize state transitions.The model combines homeostatic somnogenic and circadian alerting inputs with orexinergic stabilization and hyperarousal bias.Source: §3.2 · pp. 4–5

3.3 Corticothalamic Spiking Neural Network

  • A recurrent 100-neuron LIF network represents cortical electrodynamics, with wake-related depolarization and NREM-related hyperpolarization producing distinct activity regimes.The network uses an RC-derived LIF formulation for multi-day tractability and synaptic currents with finite rise and decay times.Source: §3.3 · pp. 5–6

3.4 Synthetic Local Field Potential and Spectral Integration

  • Synthetic LFP is formed by convolving network spiking with a dendritic response kernel, then windowed and transformed to integrate power across five canonical clinical frequency bands.The modeled bands are Delta, Theta, Alpha, Sigma, and Beta; relative power is normalized against total spectral power.Source: §3.4 · p. 7

4 Cohort Parameterization and Disorder Phenotypes

The framework maps clinical pathophysiological deficits onto biophysical parameters for five simulated cohorts, from healthy control to four sleep disorders.

Cohort 0: Healthy Normal Control

  • The healthy control uses synchronized circadian phase, intact orexin tone and cortical inhibition, and flip-flop parameters satisfying the bistability criterion.Its REM entry latency is set to 70.0 min.Source: §4 · p. 8

Cohort 1: Delayed Sleep-Wake Phase Disorder (CRSWD)

  • Delayed-phase CRSWD is modeled with a +4.0 h circadian phase offset and an approximately 50% reduction in circadian amplitude.The parameterization represents delayed endogenous phase and attenuated circadian pacemaking amplitude.Source: §4 · p. 8

Cohort 2: Chronic [Psychophysiological] Insomnia

  • Chronic insomnia is modeled with +0.60 pA arousal bias and reduced VLPO inhibitory efficacy, leaving the sleep attractor shallow and vulnerable to awakenings.The weakened switch remains above the stated critical bifurcation point.Source: §4 · p. 8

Cohort 3: Paradoxical Insomnia (Sleep State Misperception)

  • Paradoxical insomnia retains control-like subcortical parameters but reduces cortical interneuron gain to 0.25, producing beta-band intrusion while preserving slow-wave gating.The parameterization represents cortical disinhibition associated with reduced GABA concentrations.Source: §4 · p. 8

Cohort 4: Narcolepsy Type 1

  • Narcolepsy type 1 is modeled with orexin tone set to 0.05, REM entry latency reduced to 5.0 min, and mildly adjusted baseline arousal.The parameterization represents hypocretin deficiency and reduced orexinergic stabilization of REM-off nuclei.Source: §4 · p. 9

5 Empirical Validation and Clinical Scoring

The model’s synthetic spectra are calibrated against a PhysioNet Sleep-EDF N3 baseline, and its clinical scoring combines AASM sleep staging with measures of sleep quantity and restoration.

5.1 PhysioNet Sleep-EDF PSG Baseline

  • PhysioNet Sleep-EDF C3-A2 stage N3 provides reference power ratios of 0.8747 Delta, 0.0796 Theta, 0.0188 Alpha, 0.0070 Sigma, and 0.0226 Beta for epoch-wise spectral MAE evaluation.The compared bands are Delta, Theta, Alpha, Sigma, and Beta.Source: §5.1 · p. 9

5.2 PSG Staging, Clinical Metrics, and the Well-Rested Index

  • Sleep stages are assigned from hypothalamic rates, homeostatic pressure, and synthetic LFP powers using AASM criteria, alongside nine PSG and restorative metrics.These include TST, SOL, WASO, cycles, SOREMPs, the Well-Rested Index, and Spectral MAE.Source: §5.2 · p. 9
  • The Well-Rested Index multiplies homeostatic recovery at S_nadir by sleep-quota attainment, with the latter based on TST relative to 22.5 h.Homeostatic recovery scales between cleared somnogens at S_nadir = 0.18 and exhaustion at S_nadir ≥ 0.75.Source: §5.2 · pp. 9–10
  • Restorative scoring prioritizes N3 and REM over N2, penalizes circadian delay or daylight sleep, nocturnal beta intrusion and WASO, and classifies outcomes by Well-Rested Index ranges.Spectral MAE is the absolute spectral power error against the PhysioNet Sleep-EDF C3-A2 N3 spectra.Source: §5.2 · p. 10

6 Results and Clinical Inferences

Across five cohorts simulated for 72 hours under the same light-dark schedule, the model reports convergence with empirical sleep spectra alongside substantial differences between sleep duration and restorative scores. Its clinical inferences connect those differences to cortical hyperarousal, circadian phase delay, and hypocretin-related instability.

6 Results and Clinical Inferences

  • The five cohorts were simulated continuously for 72 hours under an identical 16-hour-light, 8-hour-dark photoperiod, with clinical telemetry summarized in Table 2.The simulations used 10,000 lux daylight and 0 lux at night.Source: §6 · p. 10, Table 2 · p. 10

6.1 Inference 1: Dissociation Between TST and Restorative Alertness

  • Paradoxical insomnia matched control TST at 21.15 h and other objective measures, yet its Well-Rested Index fell from 86.4% to 37.6% as beta activity persisted through N3 sleep.The model’s subjective wakefulness meter remained elevated at 85%–95% during objective slow-wave sleep.Source: §6.1 · p. 10, Figure 4 · p. 13

6.2 Inference 2: Circadian Desynchrony and the Phase-Shifted Clearance Deficit

  • A +4.0 h CRSWD phase shift delayed the circadian nadir to 10:00; sleep extended into daylight, accompanying a 47.9% Well-Rested Index, SOL of 99 min, and WASO of 454 min.The model also reports incomplete homeostatic clearance, with S_nadir ≈ 0.38 versus 0.18 in controls.Source: §6.2 · p. 11, Figure 2 · p. 11

6.3 Inference 3: Loss of Bistable Boundary Hysteresis in Hypocretin Deficiency

  • With hypocretinergic tone at Ωtone = 0.05, narcolepsy type 1 produced WASO of 781 min and 35 SOREMPs as the flip-flop switch lost hysteresis.Daytime naps partially discharged homeostatic pressure, limiting unbroken nocturnal slow-wave sleep.Source: §6.3 · p. 12, Figure 3 · p. 12

6.4 Inference 4: Spectral Convergence Against PhysioNet Sleep-EDF

  • Across all five cohorts, synthetic LFP spectra converged with PhysioNet Sleep-EDF spectra at MAE = 0.3349–0.3449, with delta dominance in N3 and beta predominance during wakefulness and paradoxical sleep.The authors interpret this correspondence as capturing physiological thalamocortical rhythms.Source: §6.4 · p. 12

7 Discussion and In Silico Pharmacology

The discussion presents DyssomniaNN as a platform for simulating targeted sleep-disorder pharmacology. It identifies limitations in spatial and REM-sleep representation and proposes pharmacokinetic modeling and patient-specific extensions.

7 Discussion and In Silico Pharmacology

  • Simulated DORAs reduce WASO in chronic insomnia, orexin agonists suppress daytime SOREMPs in narcolepsy, and GABA-A modulation suppresses beta intrusions in paradoxical insomnia.The modeled interventions respectively dampen hyperarousal, restore flip-flop bistability, and improve perceived restedness without altering subcortical macro-architecture.Source: §7 · p. 13

7.1 Limitations

  • The current corticothalamic network is a single-channel C3-A2 representation, limiting spatial coverage and leaving REM microarchitecture for future SLD and vlPAG modeling.Proposed spatial extensions would capture traveling slow waves and topographic spindle gradients.Source: §7.1 · p. 14

7.2 Future Work

  • Current simulated drug effects are static parameter shifts; adding PK/PD could represent absorption, distribution, clearance, and half-lives for comparing formulations and predicting efficacy and residual sedation.The proposed extensions include virtual comparisons of immediate- versus extended-release hypnotics.Source: §7.2 · p. 14
  • Fitting physiological parameters to wearable, actigraphy, and home-EEG data could support patient-specific digital twins and virtual testing of tailored interventions.Proposed interventions include chronotherapeutic schedules and targeted pharmacotherapies before clinical regimens are initiated.Source: §7.2 · p. 14

8 Conclusion

DyssomniaNN models sleep regulation across molecular, subcortical, and cortical mechanisms and reproduces sleep-pathology signatures across a 72-hour photoperiod. Its simulations show that sleep duration can diverge from restorative quality, while its framework supports computational evaluation of targeted interventions.

8 Conclusion

  • DyssomniaNN couples Borbély’s homeostatic and circadian processes, Saper’s hypothalamic flip-flop, and a recurrent corticothalamic LIF network across a standardized 72-hour photoperiod.The framework reproduces multi-day electrophysiological and architectural signatures of diverse sleep pathologies.Source: §8 · p. 15
  • Paradoxical insomnia combines normal sleep architecture and 21.15 h of sleep with R = 37.6% restoration, while narcolepsy type 1 combines WASO = 781 min with 26.18 h of aggregate sleep.The passage attributes paradoxical-insomnia impairment to cortical disinhibition and beta intrusions, and narcolepsy fragmentation to loss of orexinergic stabilization.Source: §8 · p. 15
  • The framework formalizes neurochemical perturbations and supports computational evaluation of targeted interventions, with spatial, PK/PD, and patient-specific extensions proposed for future work.Examples include dual orexin receptor antagonists and orexin agonists.Source: §8 · p. 15

A.1 List of Abbreviations

The appendix includes a comprehensive summary of medical, physiological, and computational abbreviations.

A.1 List of Abbreviations

  • Table 3 is a comprehensive reference summary of medical, physiological, and computational abbreviations.The supplied passage identifies the table’s scope but does not list individual abbreviations.Source: Table 3 · p. 18

A.2 List of Neurotransmitters and Neuromodulators

The appendix includes a reference guide to neurotransmitters and neuromodulators in sleep-wake neurobiology.

A.2 List of Neurotransmitters and Neuromodulators

  • Table 4 is a reference guide to key neurotransmitters and neuromodulators in sleep-wake neurobiology.The supplied passage identifies the guide’s subject but does not enumerate its entries.Source: Table 4 · p. 19
Loading 2610.13918v1…