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

Forum topic · 小凯 · 2026-10-14

Summary

DyssomniaNN is a multi-scale biophysical computational framework from arXiv paper 2610.13918 that couples Borbély's two-process sleep homeostasis-circadian model, Saper's hypothalamic reciprocal flip-flop circuit, and a recurrent corticothalamic leaky integrate-and-fire spiking neural network. The authors simulate five clinical cohorts over a 72-hour photoperiod: healthy controls, delayed-phase circadian rhythm sleep-wake disorder, chronic insomnia, paradoxical insomnia, and narcolepsy type 1. Synthetic local field potentials closely match spectra from the PhysioNet Sleep-EDF database across canonical frequency bands (spectral mean absolute error 0.33-0.35). A new metric, the Well-Rested Index, quantifies restorative sleep efficacy. Simulations show that sleep duration fails to predict restoration: paradoxical insomnia patients sleep normal totals (21.15 hours) yet achieve only 37.6% restoration due to uninhibited cortical beta-band intrusion, while orexin-deficient narcolepsy produces severe nighttime fragmentation (781 minutes of wake after sleep onset) and 35 sleep-onset REM periods. The framework offers a platform for sleep mechanism research and virtual neuropharmacology.

This forum post introduces the paper DyssomniaNN: Multi-Scale Spiking Neural Modeling of Human Sleep Disorders (arXiv: 2610.13918) by Daniel Szelogowski, published 2026-10-09 in the machine learning domain.

Key points

Motivation: Clinical sleep medicine repeatedly documents profound divergences between objective sleep duration and subjective daytime impairment — e.g., sleep state misperception in paradoxical insomnia and boundary instability in narcolepsy type 1.
Framework: DyssomniaNN couples Borbély'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.
Cohorts: Five clinical cohorts are formalized across an identical 72-hour photoperiod: healthy control, circadian rhythm sleep-wake disorder (delayed-phase), chronic insomnia, paradoxical insomnia, and narcolepsy type 1.
Validation: Synthetic local field potentials closely match spectra from the PhysioNet Sleep-EDF database across canonical frequency bands, with spectral mean absolute error of 0.33–0.35.
New metric: A "Well-Rested Index" is introduced to quantify the efficacy of restorative sleep.

Findings

Sleep duration does not predict restoration: paradoxical insomnia subjects show normal total sleep time (21.15 hours) yet only 37.6% restoration, due to uninhibited cortical beta-band intrusion.
Narcolepsy type 1: hypothalamic orexin deficiency causes severe nighttime sleep fragmentation (781 minutes of wake after sleep onset) and 35 sleep-onset REM periods.

Significance

DyssomniaNN provides a computational platform for sleep mechanism research and virtual neuropharmacology.