{"id":"f0958d3b-b990-428e-9842-cde8dd8662ba","arxiv_id":"2607.07186","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":7,"one_line_summary":"A barrier-flattening Hamiltonian replica-exchange method accelerates Mg²⁺ inner-shell binding to RNA by orders of magnitude, and a local cryo-EM reweighting framework validates individual binding motifs site-by-site.","lead":"This paper develops a computational method to dramatically speed up simulations of magnesium ions binding to RNA, then validates the results against cryo-EM maps. It matters because magnesium-RNA interactions are critical for RNA folding and catalysis but have been too slow to simulate and too ambiguous to validate experimentally.","discovery_kind":"unclear","skeptic_critique":{"model":"glm-5.2","headline":"Local reweighting cannot detect conformational distortions within motif states if the UpU-parameterized bias transfers imperfectly to the ribozyme context.","rationale":"The reader correctly identified the bias transferability assumption as the weakest link. I sharpen it by noting that the local reweighting framework — a key novel contribution — inherits this assumption silently: it can adjust motif populations but cannot detect or correct distortions in conformational statistics within each motif state caused by imperfect bias transfer. The paper's design choices (bias zero in bound/unbound states, unbiased replica used for analysis, consistency of reweighted occupancies across force fields) provide meaningful mitigation. The cryo-EM global cross-correlation analysis (Section 1.4) offers partially independent validation. The paper is also transparent about the moderate effect sizes and non-robustness of rankings to alignment choices. These factors keep the concern from being decisive: it identifies a real residual risk but not an internal inconsistency or a claim that overreaches the evidence. The verdict remains ACCEPT because the method is sound in principle, the evidence is sufficient for the claims as stated, and the limitations are honestly reported. The proposed concrete test would strengthen confidence by directly verifying that conformational statistics within motif states are preserved across biased and unbiased setups, which is the implicit assumption underlying the local reweighting framework.","tokens_in":45183,"tokens_out":6317,"duration_ms":491800,"concrete_test":"For motifs with high occupancy (~1.0) in both µMg-REx (unbiased replica) and nMg-MD, compare the distributions of Mg²⁺–O bond lengths, O–Mg²⁺–O coordination angles, and first-shell water positions. If these distributions are statistically indistinguishable across setups, the bias is not distorting bound-state conformational statistics. If they differ beyond force-field expectations, the bias transfer is introducing artifacts that the local reweighting cannot detect.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's local reweighting framework (Section 1.6) separates frames where a motif is present from those where it is absent, computes neighborhood density maps for each state, and finds the population mixture maximizing cryo-EM correlation. This procedure adjusts only the *population* of each motif independently; it cannot correct for distortions in the *conformational statistics* within the present or absent states. The conformational statistics within each state are determined by the force field and, critically, by the barrier-flattening bias U_flat, which is parameterized on the minimal UpU system and transferred to the ribozyme without reparameterization (Section 1.2: 'the barrier-flattening bias for monodentate OP1/OP2 binding generalizes to higher denticities without reparameterization'). The paper argues the barrier shape is 'only marginally affected by the local environment,' and the bias is designed to be zero in bound/unbound states, which partially mitigates this concern. However, in the folded ribozyme, steric constraints, electrostatic preorganization from nearby phosphates, and multi-phosphate coordination geometries could shift the barrier location or shape relative to UpU. If the bias does not flatten the correct barrier at a given site, it could distort the transition-state ensemble and, through detailed balance, the relative populations of substates within the 'present' or 'absent' ensembles. The cryo-EM validation at ~2.2 Å resolution provides only weak sensitivity to such distortions: the paper itself finds that force fields predicting 70 vs 45 inner-bound ions are 'largely indistinguishable,' suggesting the maps cannot resolve subtle coordination-geometry differences. Thus the validation framework may be insensitive to exactly the class of errors that bias non-transferability would introduce. This is load-bearing because the local reweighting is presented as site-by-site validation of motif populations, but its correctness silently assumes the","agreement_with_reader":"agree"},"referee_report":{"model":"glm-5.2","summary":"This manuscript presents an enhanced-sampling strategy for accelerating Mg²⁺ inner-shell binding to RNA, combining a barrier-flattening bias (designed on a model diuridine system) with Hamiltonian replica exchange. The method is applied to the Tetrahymena group I intron ribozyme, and results are validated against cryo-EM density maps using a local reweighting framework that infers individual binding-motif populations. The central findings are that insufficient sampling of inner-shell binding degrades agreement with experiment, while differences between the µMg and nMg force fields are not resolvable at current cryo-EM resolution. The work builds on the authors' top-ranked CASP16 submission and provides a methodologically detailed, internally consistent study.","tokens_in":45897,"tokens_out":1512,"duration_ms":197360,"significance":"The methodological contribution is substantial: the barrier-flattening bias is elegantly designed to be zero in bound/unbound states (enabling straightforward reweighting), generalizes to polydentate motifs without reparameterization, and is efficiently implemented with neighbor lists. The local cryo-EM reweighting framework is a pragmatic solution to the intractability of global reweighting in large systems. The systematic comparison of force fields, sampling strategies, and restraint strengths — with proper statistical treatment including Bayesian bootstrap confidence intervals — sets a high standard. The code and input files are publicly available, enhancing reproducibility. The finding that cryo-EM at ~2.2 Å resolution cannot distinguish µMg from nMg is an important negative result for the field, clearly delineating the current limits of experimental validation.","major_comments":[{"comment":"Section 1.2 and Section 1.6 (also Fig. 5): The barrier-flattening bias U_flat is parameterized on the minimal UpU system and transferred to the full ribozyme without reparameterization. The paper states that the barrier shape is 'only marginally affected by the local environment' and that the bias is zero in bound/unbound states. While the zero-in-states property is a genuine mitigation, the local cryo-EM reweighting framework (Section 1.6) adjusts only the population of each motif independently — it cannot correct for distortions in the conformational statistics within the 'present' or 'absent' ensembles caused by an imperfectly transferred bias. In the folded ribozyme, steric constraints, electrostatic preorganization, and multi-phosphate coordination could shift the barrier location or shape relative to UpU. The paper would be strengthened by a more direct test of transferability: for","section":null},{"comment":"Section 1.4, Fig. 4C: The global cryo-EM validation shows that the ranking of setups (e.g., nMg-MD vs. µMg-REx) is 'not robust to a change in alignment and voxel selection,' and that the strongest factor is the restraint strength on RNA. The authors acknowledge that 'observed variations in density are dominated by how cations organize RNA local dynamics rather than direct features of the solvation shell.' This raises a concern about whether the cryo-EM validation is actually testing ion sampling or primarily testing RNA conformational restraint protocol. The Core Phosphates Shell analysis (excluding RNA itself) is the most direct test of ion placement, but even there the differences between µMg-REx and µMg-MD remain below significance. The authors should more explicitly state the limitations of cryo-EM validation at this resolution for distinguishing sampling strategies, and clarify that","section":null},{"comment":"Section 1.6, Fig. 5 and Table S5: The local reweighting procedure computes, for each motif, the population mixture maximizing cross-correlation with the experimental map. This is a fitting procedure with one free parameter per motif. The paper frames it as 'integration' rather than 'prediction,' which is appropriate. However, the risk of overfitting is not discussed. With ~60+ motifs each fit independently, some will achieve high CCmax by chance. The CCmax < 0.7 threshold (Fig. S12) removes poorly resolved regions but does not control for false positives among the remaining sites. A brief discussion of this limitation — and whether the agreement across independent ensembles (µMg-MD, nMg-MD, µMg-REx) provides a natural cross-validation — would strengthen the analysis. The cross-ensemble agreement is noted ('reweighted occupancies agree across conformational ensembles') but could be quant","section":null}],"minor_comments":[{"comment":"Section 1.4: The strong conformational restraints on RNA (κ = 10^5 kJ/mol/nm²) are a significant methodological choice that limits RNA flexibility. While the authors acknowledge this and test a softer restraint (κ = 5×10²), the soft-restraint results perform significantly worse. It would help to briefly discuss whether the strong restraints might artificially enforce ion-binding geometries that depend on the native fold, particularly for keystone motifs.","section":null},{"comment":"Figure 5: The figure is information-dense and difficult to parse in its current form. The four highlighted motif examples with local cross-correlation plots are useful, but the main heatmap of motif frequencies across setups is hard to read. Consider splitting into separate panels or providing an interactive version.","section":null},{"comment":"Section 1.1: The interchange associative (Ia) mechanism observed for µMg is noted to be 'at odds' with the Id or D mechanism suggested by experiment and ab initio calculations. This discrepancy is mentioned but not discussed further. A brief comment on whether this affects the reliability of the bias design would be helpful.","section":null},{"comment":"Table S1: The 'indicative' performance numbers (h/ns) and total CPU time are useful but the footnote notes actual setups 'may have differed.' For reproducibility, providing the exact wall-clock times and hardware used for each setup would be preferable.","section":null},{"comment":"Section 3.2: The paper mentions a PLUMED bug fix for the cosine switching function that was backported during the study. This is good practice to report, but it would help to clarify whether any results in the paper were affected by the pre-fix version of the code.","section":null},{"comment":"The reference list includes several 2025–2026 dates (e.g., refs 11, 12, 15, 17, 37, 63), suggesting these are preprints or in-press articles. The authors should verify that all citations are to published or properly versioned preprints at submission.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The stress-test concern about bias transferability from UpU to the ribozyme is legitimate but, in my assessment, does not rise to a major revision. The bias is designed to be zero in the relevant end states, the Kish ratio analysis shows good reweighting efficiency, and the cross-ensemble agreement in the cryo-EM reweighting provides indirect evidence that the bias is not severely distorting the ensemble. The concern about the local reweighting not detecting within-state conformational distortions is theoretically valid but is a known limitation of any reweighting approach at finite resolution, and the authors are appropriately cautious in their claims. I have raised it as a major comment requesting more discussion and ideally a direct test, but it does not undermine the central claims of the paper."},"author_rebuttal":null,"desk_editor":{"model":"glm-5.2","letter":"The main thing to know: this paper introduces a genuinely new combination of a hand-crafted barrier-flattening bias on coordination-number CVs with Hamiltonian replica exchange to accelerate Mg²⁺ inner-shell binding to RNA, plus a local motif-by-motif cryo-EM reweighting framework that infers individual binding-site populations from density agreement. Both are real contributions to a recognized bottleneck in RNA structural biology. Code and data are public (GitHub link provided), which matters here because the method has moving parts that need reproducing to trust fully. The CASP16 top-ranking context is legitimate background, not self-promotion — the current paper is a substantially refined version of that preliminary submission, with new analysis and new validation framework. The UpU model-system characterization is thorough: free-energy landscape, rate acceleration quantified with confidence intervals, Kish ratio analysis showing >80% weighting efficiency, and honest comparison to the faster nMg model. The ribozyme application includes proper controls (no-Mg, soft restraints, novo vs. PDB-informed, multiple replicates). The finding that sampling dominates over force-field choice at current cryo-EM resolution is well-supported and honestly reported — they don't oversell what the maps can resolve. The local reweighting framework is the more novel analytical piece and is correctly framed as inference, not prediction. The stress-test concern about bias transferability from UpU to the ribozyme is real but probably overstated. The bias is designed to be zero in bound and unbound states, acting only near the barrier saddle point, and the cryo-EM validation does provide independent evidence that the approach works — 12 of 16 known motifs recovered de novo. That said, the concern that local reweighting adjusts populations but not within-state conformational distortions is technically correct and worth flagging in review. If the barrier geometry differs in the folded context due to steric or electrostatic effects, the bias could distort transition-state ensembles in ways the ~2.2 Å maps cannot detect. This is a genuine limitation but not a fatal one — the paper acknowledges the assumption explicitly, and the weight of evidence supports reasonable transferability. The soft spots are minor: the free-parameter count is moderate but defensible given the systematic B-factor and cutoff scans, and the RNA restraints are strong (κ = 10⁵ kJ/mol/nm²), which the authors acknowledge may obscure force-field differences. This paper is for method developers in RNA/biomolecular simulation and structural biologists working with cryo-EM on RNA systems. It deserves a serious referee who can evaluate the PLUMED implementation details and the reweighting math. Recommend accept for peer review.","headline":"Barrier-flattening HREX for Mg²⁺–RNA binding plus local cryo-EM motif reweighting — solid method paper with one load-bearing transferability assumption","tokens_in":46297,"tokens_out":638,"would_cite":true,"duration_ms":99992,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"glm-5.2","headline":"Sampling beats force fields for RNA magnesium binding","keywords":["magnesium","RNA","enhanced sampling","cryo-EM","molecular dynamics","barrier flattening","replica exchange","binding motifs"],"falsifier":"If a future cryo-EM map at substantially higher resolution (e.g., sub-1.5 Å) showed that one force-field model systematically matches experimental inner-shell versus outer-shell populations while the other does not, the claim that force-field differences are undetectable would be falsified. Additionally, if the barrier-flattening bias were found to produce qualitatively wrong motif populations at sites where the local environment substantially alters the transition-state geometry, the transferability assumption would fail.","tokens_in":45441,"feed_emoji":"🧲","tokens_out":1207,"duration_ms":185764,"temperature":0.7,"pith_summary":"Magnesium ions bind RNA at specific sites through a slow process where the ion must shed a water molecule from its tightly held inner coordination shell and replace it with an RNA atom. This exchange happens on microsecond timescales, far beyond what standard molecular dynamics simulations can reach. The paper builds a two-part solution: a hand-crafted energy bias that flattens the barrier for this water-to-RNA swap, enabling sub-nanosecond binding events, combined with Hamiltonian replica exchange to apply this acceleration across hundreds of equivalent binding sites simultaneously in a large ribozyme. The key empirical finding is that when this accelerated sampling is compared against experimental cryo-EM density maps, the dominant factor determining agreement with experiment is whether inner-shell binding was adequately sampled at all. Two different magnesium force-field models that predict measurably different ratios of inner-bound to outer-bound ions produced nearly identical agreement with the experimental maps, because current cryo-EM resolution cannot distinguish between those binding modes. The paper also introduces a site-by-site analysis method that uses local cryo-EM density to infer the population of each individual binding motif independently, circumventing the exponential loss of statistical power that would plague a global reweighting of hundreds of simultaneous binding sites.","feed_headline":"Sampling beats force fields for RNA magnesium binding","feed_subtitle":"Accelerated simulations reveal that adequate sampling, not force-field choice, determines whether predicted magnesium binding sites match","key_machinery":"The barrier-flattening bias U_flat acts on a two-dimensional space defined by the magnesium ion's coordination number with water oxygen atoms and with RNA phosphate oxygen atoms. It is a crescent-shaped potential that is zero in both the bound and unbound states, applying force only near the transition-state saddle point where the ion is heptacoordinated. Because all magnesium ions are chemically equivalent and the barrier shape is approximately environment-independent, the same bias is applied to every ion simultaneously via Hamiltonian replica exchange across a ladder of eight replicas with increasing bias strength. The local cryo-EM analysis separates simulation frames where a given motif","core_discovery":"The central discovery is a quantitative separation of concerns: for modelling magnesium binding to RNA, inadequate sampling of inner-shell binding is the primary bottleneck limiting agreement with experiment, while differences between current force-field parameterizations are secondary and largely undetectable at available experimental resolution. The method that establishes this combines a transferable barrier-flattening potential (parameterized once on a minimal two-nucleotide system and applied without reparameterization to polydentate sites in a 386-phosphate ribozyme) with a local cryo-EM integration framework that validates individual binding motif populations rather than requiring a全局","pith_inferences":["If the barrier-flattening bias is truly environment-independent, one could construct a universal library of such biases for different ion-ligand chemistries (e.g., Ca2+, Mn2+, Zn2+) and apply them as modular plug-ins to any biomolecular simulation.","The local reweighting approach could be extended into an iterative refinement loop: use cryo-EM-inferred populations to reseed simulations, re-sample, and re-validate, potentially converging to ensembles that are self-consistent with experiment without global reweighting.","The observation that RNA conformational restraints dominate the cryo-EM cross-correlation suggests that current RNA force fields may have larger systematic errors in conformational dynamics than in ion binding, implying that ion-binding accuracy is limited not by the ion model but by the RNA model.","If cryo-EM resolution improves beyond the current 2.2 Å, the framework could be used to discriminate between force-field models, turning this validation tool into a force-field selection criterion."],"forward_implications":["If sampling is the dominant bottleneck, then existing RNA structures in the PDB with manually placed magnesium ions may contain systematic errors from under-sampled binding modes that enhanced sampling could correct.","The local reweighting framework is generalizable to any system where many weakly coupled local events contribute to an experimental observable, including protein-ligand binding, lipid flip-flop, or proton transfer in reactive dynamics.","The inability of cryo-EM to distinguish inner-shell from outer-shell magnesium binding means that ion-counting experiments or X-ray methods providing different observables would be needed to discriminate between force-field models.","The finding that a bias parameterized on a minimal model transfers to the full ribozyme suggests that local free-energy barrier shapes are robust to structural context, which if general would simplify enhanced-sampling design for other biomolecular systems."],"fun_headline_variants":["Sampling, not force fields, limits RNA magnesium modeling","Cryo-EM and accelerated sampling resolve RNA magnesium sites","Barrier-flattening simulations expose RNA magnesium binding","Inadequate sampling dominates errors in RNA magnesium models","Transferable enhanced-sampling method validates RNA Mg²⁺ sites"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The barrier-flattening bias was designed and fitted on a minimal system of one magnesium ion with a two-nucleotide fragment, then transferred to the full ribozyme without reparameterization. This assumes the free-energy barrier shape for water-to-phosphate exchange is essentially the same regardless of local structural context, steric constraints, or electrostatic environment in the folded RNA. If the barrier geometry differs in the folded ribozyme, the bias could distort the","fun_headline_variants_meta":{"raw":{"variants":["Sampling, not force fields, limits RNA magnesium modeling","Cryo-EM and accelerated sampling resolve RNA magnesium sites","Barrier-flattening simulations expose RNA magnesium binding","Inadequate sampling dominates errors in RNA magnesium models","Transferable enhanced-sampling method validates RNA Mg²⁺ sites"]},"model":"glm-5.2","effort":"low","cost_usd":0.0,"raw_usage":{"total_tokens":599,"prompt_tokens":521,"completion_tokens":78,"prompt_tokens_details":null},"tokens_in":521,"tokens_out":78,"duration_ms":15147,"temperature":1.0,"reasoning_tokens":null,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-09T17:50:51.043998+00:00","model_set":{"reader":"glm-5.2"},"falsifier":"If a future cryo-EM map at substantially higher resolution (e.g., sub-1.5 Å) showed that one force-field model systematically matches experimental inner-shell versus outer-shell populations while the other does not, the claim that force-field differences are undetectable would be falsified. Additionally, if the barrier-flattening bias were found to produce qualitatively wrong motif populations at sites where the local environment substantially alters the transition-state geometry, the transferability assumption would fail.","supporting_citations":[],"review_version":1}