The NMDA receptor is a heterotetramer, typically assembled from two GluN1 and two GluN2A subunits, and it demands two co-agonists before it will open: glycine binds GluN1, glutamate binds GluN2A, and the magnesium block must be relieved. Once activated, the receptor undergoes a cascade of conformational changes that dilate a central pore, allowing calcium and sodium to pour into the cell. Because mutations in this receptor are linked to Alzheimer’s disease, depression, stroke, epilepsy, and schizophrenia, understanding the open state is not merely an academic exercise; it is a prerequisite for rational drug design. Yet the open state is transient, heterogeneous, and notoriously difficult to capture experimentally, which is precisely why the Buffalo team, led by Wenjun Zheng with first author Xing Liu, turned to simulation.

The researchers built a truncated model of the GluN1/GluN2A receptor containing only the ligand-binding domains and the transmembrane domain, the two layers that carry the essential channel-gating machinery. They started from a crystal structure of an allosterically inhibited GluN1/GluN2B receptor, generated a homology model for the rat GluN1 and GluN2A sequences using SWISS-MODEL, embedded the truncated receptor in a POPC lipid bilayer with CHARMM-GUI, and then drove it toward an active-like conformation with targeted molecular dynamics, using an active-state ligand-binding domain structure derived from a cryo-EM map as the target. Once there, they held the ligand-binding domains in their active conformation with positional restraints and let the transmembrane domain move freely across seven independent 800-nanosecond trajectories at 30 degrees Celsius.

The resulting ensemble revealed something striking: the channel pore is not a static tunnel but a violently fluctuating one. Analysis with the HOLE program showed the minimal pore radius swinging between 0 and 2.8 angstroms across the simulation frames. Using established criteria for ion-channel conductance, the team estimated that 55 percent of frames were water permeable, 35 percent partially ion-conducting, and less than 1 percent fully open. The authors interpret these rapid pore-radius fluctuations as a dynamic feature of the transmembrane domain in the active-like state, distinct from the much slower transitions between fully open and closed functional states. In other words, even when the receptor is nominally active, its gate breathes.

Zooming in on individual residues, the team calculated root mean square fluctuations across the ensemble and found pronounced flexibility in the S1-M1 linker, the M1-M2 loop, the M2-M3 loop, the M3-S2 linker, the S2-M4 linker, and the C-terminus of the M4 helix. Crucially, the two subunit types behaved differently. GluN1 showed higher flexibility in the S1-M1, M3-S2, and S2-M4 linkers, while GluN2A was more mobile in M1 and the M2-M3 loop. At the two proposed gates, the upper gate at the bundle crossing of the M3 helices and the lower gate at the apex of the M2 pore loops, the fluctuations were moderate but subtly asymmetric between subunits, hinting that GluN1 and GluN2A play distinct roles in controlling channel opening and closing.

Principal component analysis of the combined trajectories identified ten dominant modes of collective motion that together account for 83 percent of the structural variation. Three of these modes correlated significantly with pore radius, and all three showed subunit-dependent amplitudes, with GluN2A moving more vigorously near the lower gate while the two subunits moved similarly near the upper gate. The analysis also highlighted the M1-M2 loop, a region so flexible that it is often missing from experimentally solved structures, as a hub of collective motion. These findings reinforce a growing consensus that the NMDA receptor’s active state is not a single conformation but an ensemble of rapidly interconverting sub-states.

To connect motion to mechanism, the researchers catalogued every hydrogen bond and van der Waals contact between residue pairs across all 5,600 simulation frames and computed Pearson correlations between each interaction and the minimal pore radius. Interactions with correlations above 0.3 or below negative 0.3 were flagged as potentially favoring opening or closing, respectively. The analysis yielded five hydrogen-bonded pairs that appear to stabilize the open pore, including a cross-subunit contact between GluN1 M4 and the GluN2A M2-M3 loop, and fifteen pairs that appear to favor closure, many clustered in the flexible linkers that transmit force from the ligand-binding domains down to the pore. In total, 35 key residues emerged from the hydrogen-bond analysis, 33 of them evolutionarily conserved and 11 known sites of disease mutations recorded in ClinVar and HGMD.

The van der Waals analysis painted an even denser picture, identifying 144 residues whose nonpolar contacts correlate with pore dynamics, 125 of them conserved and 49 linked to disease. Because so many transmembrane residues are conserved and therefore plausibly functional, the correlation-based approach produced a long list of candidates that was too diffuse to guide experiments efficiently. To sharpen the predictions, the team turned to machine learning. They trained a random forest regressor to predict the minimal pore radius from pairwise interaction energies and hydrogen-bond occupancies, then used SHAP values, an explainable-AI technique borrowed from cooperative game theory, to rank the importance of every interaction. The result was a much sparser list: just 32 residues in the top 10 percent of importance, 29 of them conserved and 19 carrying disease mutations, an enrichment from under 25 percent to roughly 60 percent compared with the correlation analysis.

Notably, twelve of those 32 machine-learning hits had not been flagged by the Pearson correlation analysis at all, suggesting that nonlinear relationships between residue contacts and pore geometry exist that simple linear statistics miss. The machine-learning model, however, came with an honest caveat: it learned a trajectory-specific relationship that failed to generalize across independently sampled trajectories, meaning the model captured local statistical structure rather than a universal physical law. The authors are candid about this limitation, as well as about the artifacts that truncating the amino-terminal domain and restraining the ligand-binding domains may introduce, and they call for larger-scale simulations of the full-length receptor as computational hardware improves.

The study’s most actionable contribution may be its bridge to medicine. By annotating predicted gating residues with evolutionary conservation grades and known pathogenic mutations, the work offers a concrete, testable hypothesis: that disease-associated variants in the S1-M1 linker, the M2-M3 loop, and the M4 helix disrupt channel gating by perturbing the transient network of polar and nonpolar interactions that stabilizes the open pore. A recent functional study independently showed that the S1-M1 linker critically controls channel opening, lending weight to this framework. The next step, the authors suggest, is to simulate specific disease mutations to determine whether they produce gain-of-function or loss-of-function phenotypes, turning a 5.6-microsecond computational experiment into a roadmap for understanding, and eventually treating, the neurological disorders that arise when this molecular gate malfunctions.

Subject of Research: Molecular dynamics simulation and machine learning analysis of the active-like state of the GluN1/GluN2A NMDA receptor channel

Article Title: Molecular dynamics simulation of the truncated NMDA receptor in a partially restrained active-like state

Article References: Molecular dynamics simulation of the truncated NMDA receptor in a partially restrained active-like state. (n.d.). https://doi.org/10.3389/fchbi.2026.1894245

Image Credits: AI Generated

DOI: 10.3389/fchbi.2026.1894245

Keywords: NMDA receptor, molecular dynamics, GluN1, GluN2A, ion channel gating, machine learning, random forest, SHAP, hydrogen bonds, van der Waals interactions, synaptic plasticity, disease mutations

The method, which the authors call Quantum Feature Selection (QFS), exploits analog quantum simulation with arrays of neutral atoms excited to Rydberg states. In these processors, individual atoms are trapped by optical tweezers and act as two-level systems, with a ground state standing for zero and a highly excited Rydberg state standing for one. Crucially, atoms in Rydberg states interact through van der Waals forces that fall off with the sixth power of distance. That fixed physical law, normally a constraint, becomes the computational resource: by rearranging the atoms in a plane, the researchers can program the interaction structure of an optimization problem directly into the geometry of the array.

The mapping works in two steps. First, the relevance of each feature, quantified by its mutual information with the target variable, is encoded as a site-dependent local detuning, a laser parameter that makes certain atoms energetically more likely to be excited. Features with high relevance receive stronger detunings, biasing their atoms toward the Rydberg state. Second, pairwise redundancy between features, measured as mutual information between the features themselves, is translated into spatial distance. The team applies an inverse-sixth-root transform to the normalized redundancy matrix, so that highly redundant pairs of features are placed close together, where their strong interaction penalizes the configuration in which both are simultaneously selected. Weakly correlated features are pushed apart, where they behave nearly independently.

Because a general redundancy matrix cannot usually be embedded exactly in two dimensions, the researchers use multidimensional scaling (MDS), a classical technique that finds atomic positions best approximating the target distances. The resulting layout is then rescaled so that the closest atom pairs sit near the Rydberg blockade radius, the critical distance at which joint excitation is suppressed. An annealing protocol follows: the global Rabi frequency ramps up and down smoothly, a global detuning initializes the system, and the local detunings encoding relevance switch on in the latter half of the evolution. The dynamics steer the array toward low-energy configurations that balance relevance against redundancy, and repeated projective measurements yield bitstrings, each one a candidate feature subset.

The post-processing stage is deliberately conservative. Only the lowest-energy ten percent of measured bitstrings are retained, and features are ranked by their average Rydberg excitation density across this ensemble. A greedy construction then accepts a feature only if its normalized redundancy with all previously accepted features falls below a threshold, set to 0.7 in the reference experiments, ensuring that the final subset is not merely relevant but internally non-redundant. When the threshold rule cannot fill a requested subset size, the remaining slots are filled by density ranking, a completion step the authors flag explicitly so that large subsets are not misread as strict independent sets.

Evaluated in simulation on Amazon Braket, using a noise-free Schrödinger equation solver and 10,000 measurement shots per run, QFS was tested on three real-world binary classification benchmarks: Adult Income, Bank Marketing, and Telco Churn. The comparison set was demanding, including mutual-information ranking, Random Forest and XGBoost feature importances, L1-regularized logistic selection, and two classical optimizers of exactly the same relevance-redundancy objective, an exhaustive search called Exact-Q and a greedy heuristic called Greedy-Q. Across ten stratified train/test splits with a Random Forest downstream classifier, QFS achieved competitive, dataset-dependent performance. On Adult Income, QFS, Exact-Q, Greedy-Q, and XGBoost importance all landed within roughly three ten-thousandths of a point in mean AUC, while on Telco Churn every method was essentially tied within split-to-split variability.

The authors are candid about where the method falls short. On Bank Marketing, tree-based importance rankings achieved slightly higher mean AUC than the quantum-derived subsets, and QFS sometimes reached its best performance at larger subset sizes, a consequence of the evaluation protocol rather than an intrinsic preference of the sampler for compact sets. The geometric embedding itself introduces moderate distortion: the mean relative reconstruction error of the MDS layouts ranged from about 0.22 to 0.26 across the three datasets. Yet a stability analysis over 50 random embedding seeds showed that this distortion does not produce arbitrary subsets. Bank Marketing was invariant across seeds for subset sizes up to five, and Adult Income stabilized for sizes of three or more, with mean Jaccard similarity above 0.96.

Robustness tests of the classical extraction stage added further reassurance. Varying the retained low-energy fraction from 5 to 30 percent shifted the best AUC only marginally on a reference split, and injecting controlled noise into the measured excitation densities, mimicking readout fluctuations or calibration drift, left predictive performance largely intact even when the exact identity of the selected subsets drifted at small cardinalities. The authors interpret this as evidence that the pipeline’s predictive output is more robust than its exact subset choices, though they stress that the simulations remain noise-free: decoherence, atom loss, finite-temperature atomic motion, readout errors, and detuning miscalibration were all excluded and must be addressed in future device-level studies.

What makes the work notable is less a claim of quantum advantage than the physical interpretability of the encoding. Every term of the statistical objective has a hardware counterpart: relevance lives in local detuning amplitudes, redundancy in the sixth-power interaction law, and the relevance-redundancy trade-off emerges from the relative energy scales of the two. The main contribution, the authors write, is a neutral-atom implementation of a relevance-redundancy feature-selection objective together with a quantitative characterization of its geometric and post-processing approximations. Scaling up will require one atom per feature, site-resolved detuning control, and embedding strategies, such as sparsification or community-aware layouts, that preserve the dominant redundancy structure without forcing every pairwise relation into a single plane.

The study points toward a broader program in which analog quantum processors tackle the combinatorial core of data preprocessing rather than the models themselves. Future work, the team says, should focus on experimental validation on real neutral-atom devices, noisy analog Hamiltonian simulation, improved weighted and graph-aware embeddings, dynamic detuning optimization, and larger benchmarks where the scaling of analog sampling can be measured directly. Extensions to multiclass problems, regression, and unsupervised settings are also on the table. For now, QFS stands as a proof of concept that a laser-driven array of atoms, arranged according to the statistical anatomy of a dataset, can select features as well as established classical heuristics, and that the path from information theory to atomic physics can be traversed without losing the meaning along the way.

Subject of Research: Analog quantum feature selection using neutral-atom Rydberg processors

Article Title: Analog quantum feature selection with neutral-atom quantum processors

Article References: Orquín-Marqués, J. J., Flores-Garrigós, C., Gomez Cadavid, A., Simen, A., Solano, E., Hegade, N. N., Martín-Guerrero, J. D., & Vives-Gilabert, Y. (2026). Analog quantum feature selection with neutral-atom quantum processors. Quantum Machine Intelligence, 8(2), Article 108. https://doi.org/10.1007/s42484-026-00456-8

Image Credits: AI Generated

DOI: 10.1007/s42484-026-00456-8

Keywords: neutral atoms, Rydberg arrays, quantum feature selection, analog quantum simulation, mutual information, quantum optimization, machine learning, multidimensional scaling, Rydberg blockade, QUBO, Amazon Braket, Quantum Machine Intelligence