Revealing Intermediate States of Voltage-Gated Potassium Channels with Physics-Informed Deep Learning: Application to Kv7.1
Kacher J., Tarek M., Frontiers in Biophysics, 2026
https://www.frontiersin.org/journals/biophysics/articles/10.3389/frbis.2026.1877588/abstract
Abstract
Voltage-gated potassium channels convert changes in membrane voltage into pore opening through conformational transitions that remain difficult to resolve as continuous pathways. Kv7.1/KCNQ1 provides a stringent benchmark for this problem because its voltage sensor populates resting, intermediate, and activated configurations, while phosphatidylinositol 4,5-bisphosphate and KCNE subunits reshape coupling between voltage-sensor motion and pore opening. Here, we present a scalable physics-informed deep learning workflow to infer Kv7.1 gating pathways from molecular dynamics ensembles of the tetrameric channel core.
A convolutional autoencoder trained only on resting/closed and activated/open endpoint ensembles, and constrained by geometric, molecular-physics, and Kv7.1-specific gating-charge terms, placed the withheld intermediate/open ensemble in an interpretable region of the latent manifold. Independent cryo-electron microscopy structures further projected along the expected down, intermediate, and up voltage-sensor sequence. Decoded intermediates followed a progressive activation signature, including an effective resting-to-activated displacement of the S4 elementary charges and sequential exchange of charge-pairing contacts. Pore analysis revealed late activation-gate widening despite the absence of direct pore-radius supervision. The generated structures retained acceptable stereochemistry and remained compatible with reinsertion into an explicit membrane environment and subsequent simulations. Applying the same workflow to IKs-derived data separated one-and two-PIP2-per-Kv7.1-monomer conditions and identified remodeling in voltage sensor, S4-S5 linker, and S6-gate contact networks.
Together, these results support physics-informed generative modeling as a route for converting sparse endpoint simulations and structural snapshots into testable, pathway-resolved atomic hypotheses for ion-channel gating.
