Interpretable Machine Learning Identifies a Multi-regional Absence Seizure Mechanism
Abstract number :
2.48
Submission category :
1. Basic Mechanisms / 1C. Electrophysiology/High frequency oscillations
Year :
2025
Submission ID :
1392
Source :
www.aesnet.org
Presentation date :
12/7/2025 12:00:00 AM
Published date :
Authors :
Presenting Author: Jacob Hull, PhD – Stanford University
Surya Ganguli, PhD – Stanford University
John Huguenard, PhD – Stanford University
Rationale: Absence seizures (AS) are characterized by synchronized spike wave discharges occurring across multiple cortical areas. Numerous mutations, brain structures, and cell types are identified as generating AS. A focus on single brain regions or ion channels may however overlook common dynamics which achieve similar effects on inter-regional communication. Due to the brain’s complexity, describing these interactions requires a mathematical framework but without prior knowledge of the equations this approach is intractable. Here we use interpretable data-driven dynamical model discovery, identifying a system of equations describing AS generation, capturing multiple phase, amplitude, and frequency-dependent interactions. Using silicon probes, we then identified the specific cortical and subcortical processes corresponding to model identified functions and used multisite optogenetics to demonstrate a causal role of these phase dependent interactions in absence seizure generation.
Methods: We used 16-site electrocorticogram (Ecog) recordings from Scn8a+/- and mouse model of absence and the sparse identification of nonlinear dynamics (SINDy) machine learning algorithm, to discover AS governing equations. We then used silicon probes (1152 sites) to identify physiological correlates over thousands of individual neurons and used multisite optogenetic stimulation of premotor and somatosensory layer 5 pyramidal neurons and the higher order posterior thalamic nucleus (PO).
Results: In our Ecog recordings, SINDy identified equations for nonlinear oscillators with phase and amplitude-dependent coupling. Simulation recapitulates 60 measures of AS oscillation phase, amplitude, and frequency relationships across regions simultaneously. Phase plane analysis identifies that seizure activity arises from phase locking driven amplitude growth driven by the coupling strength between somatosensory and motor related regions. In our high density silicon probe recordings we identify an AS onset network between layer 5 pyramidal neurons of somatosensory and premotor cortex and the higher order posterior thalamic nucleus (PO) which corresponds to the model identified coupling functions. Optogenetic stimulation of PO and premotor layer 5 neurons reveal a phase difference dependent enhancement of layer 5 pyramidal neuron bursting vs cortical inhibitory neuron recruitment. Combined premotor and somatosensory layer 5 pyramidal neuron optogenetic stimulation then revealed a phase dependent cooperative AS induction.
Conclusions: Our interpretable machine learning approach identifies a quantitative mechanism of AS generation resulting from thalamic, specifically PO, enhancement of long range oscillatory cortico-cortico communication, providing multiple therapeutic targets which are not apparent by studying these brain regions in isolation.
Funding: 5R01NS117150 to JRH and SG, 1F32NS123009 to JMH
Basic Mechanisms