{"id":"7fdb3f62-49ba-4a07-a2d8-658c27388372","arxiv_id":"2509.20560","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Symmetric training of mechanical networks produces cooperative binding, and a modified training that raises the frequency of the functional mode increases the temperature at which thermal fluctuations destroy that cooperation.","lead":"The paper trains model elastic networks to respond symmetrically between two binding sites, showing this gives cooperative binding, and it derives a temperature above which thermal motion breaks the response. It then modifies the training to stiffen the relevant vibrational mode, lifting that temperature so the function survives at biologically relevant temperatures.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"300-K claim depends on an upward-biased binding-strain estimate (SI p*≈83) and a lenient 1σ crossover; either correction may remove the Fig. 4 range.","rationale":"Read in good faith, the paper's internal machinery is convincing: Eq. 6 is standard linear-response variance, the MD validation (SI Figs. S1–S2) shows 6% average error up to T*, and the r-modified training demonstrably raises Θ* in a nonmonotonic way consistent with stiffness saturation. The central risk is not internal consistency; it is the bridge from dimensionless network temperatures to 'biologically relevant' temperatures. The reader's weakest assumption pointed to the energetic origin of cooperativity; I agree that is a limitation, but the more immediately load-bearing assumption is quantitative: the 300-K threshold in Figs. 2–4 is placed using ε_s≈0.19, derived in the SI by optimizing a percentile p* of pairwise atom strains. That procedure is designed to filter out 'uninformative' small strains and 'extreme' large strains, but nothing guarantees the resulting p* corresponds to the strain of a coarse-grained network bond. Because T* scales as ε_s^{-2}, a factor of 2 in ε_s changes the threshold by 4. Similarly, the 1σ criterion in Eq. 8 is a permissive definition of 'crossover'; if functional robustness means 2–3σ distinguishability, T* shifts down by 4–9. Either shift is large compared to the width of the gray band and could remove the Fig. 4 window. I also note the acknowledged energy-only mechanism (Discussion) and the unreported training failure rate, but these are secondary to the quantitative hinge. The proposed reanalysis with alternative ε_s and σ-criterion is a single computational check that would settle whether the headline biological claim survives.","tokens_in":21854,"tokens_out":14488,"duration_ms":111642,"concrete_test":"Recompute the circled points in Fig. 4 under two single changes: (1) replace ε_s=0.19 with ε_s=0.09 (the authors' 25th percentile) in the conversion for T*, leaving TM unchanged; (2) keep ε_s=0.19 but change Eq. 8 to ⟨εt⟩ = n√(k_B T*)\tildeσ with n=2 (and optionally n=3). If either change removes all circled (Δ,r) points, the 300-K claim is not robust to reasonable parameter/criterion choices; if circled points survive, the concern is settled.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim—that there is a (Δ,r) window with both crossover and melting temperatures above 300 K (Fig. 4)—rests on the dimensionless-to-real temperature conversion in Sec. VII C and the SI's estimation of the binding strain εs. The conversion for T* uses Θ*/(ε_s^2) with ε_s in the denominator squared (Eq. 8 and SI Eq. 3). The authors select ε_s through a score-maximizing percentile p*≈83 of intra-protein atom-pair strains (SI, 'Median Value for Source Strain εs'), yielding M(p*)≈0.19. This is not an obvious physical choice: a coarse-grained network bond should represent a typical interface deformation, not the 83rd percentile of pairwise atom displacements. Their own 25th–75th percentile range at p* is 0.09–0.31, so using 0.09 raises the 300-K threshold by (0.19/0.09)^2≈4.5; using the raw per-pair median before percentile filtering (Fig. S17) would raise it far more. A separate but compounding choice is the crossover definition itself: Eq. 8 sets T* at 1σ separation of bound and unbound target-strain distributions. If a 2σ or 3σ separation is required for 'robust' function, T* drops by a factor of 4 or 9. Either correction could eliminate the circled points in Fig. 4. This does not affect the internal mechanics—the linear-response formula and MD validation are solid—but it determines whether the headline 'biologically relevant temperatures' is actually supported. The energy-only limitation stated in the Discussion is real but secondary; even granting it, the mapping is the load-bearing hinge.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper trains disordered elastic spring networks at zero temperature for cooperative (bidirectional) allostery, defines a crossover temperature T* above which thermal fluctuations destroy the trained response, and introduces a modified contrast function with parameter r that raises T* by stiffening the relevant low-frequency mode. The authors validate their linear-response strain-fluctuation calculation against LAMMPS simulations, compute a Lindemann-type melting temperature, and compare both T* and T_M to a 300 K threshold using protein-derived estimates of strain, length, and stiffness. They conclude that for a range of (Δ, r) both temperatures lie above biologically relevant values, so cooperative function can be thermally robust.","tokens_in":22287,"tokens_out":4699,"duration_ms":37662,"significance":"The linear-response framework (Eqs. 6, 17) is standard and the validation against MD in the SI (average ~6% error up to T*) is a real strength. The idea that training for bidirectional allostery yields cooperativity, and that the associated soft mode controls thermal robustness, is plausible and potentially useful for designing synthetic mechanical networks. The modified training protocol with r is a clean extension. However, the headline claim of robustness at 300 K depends on a chain of dimensional conversions and criterion choices that are not fully justified; the internal mechanics are sound, but the biological extrapolation is sensitive to the chosen binding strain and to the 1-sigma crossover definition. The authors explicitly acknowledge that their mechanisms are energetic and not entropic, which is a genuine limitation for protein relevance.","major_comments":[{"comment":"The central 300 K comparison in Fig. 4 depends on the representative source strain εs through Eq. 8, where T* scales as ε_s^2. The SI selects εs=0.19 by maximizing U(p)=M(p)/(W(p)+1/M'(p)), yielding p*≈83% of the intra-protein atom-pair strain distribution. This is not a physical argument that a coarse-grained bond strain maps to the 83rd percentile of atom-pair strains. The authors' own 25th–75th percentile range is 0.09–0.31; using 0.09 increases the 300 K threshold by (0.19/0.09)^2≈4.5, which can remove the circled points in Fig. 4. Please provide a robustness analysis or a more principled determination.","section":"SI, 'Median Value for Source Strain εs'; Fig. S19"},{"comment":"Eq. (8) defines T* as the temperature at which the 1-sigma unbound strain fluctuation equals the mean bound strain. Calling this 'the crossover temperature above which functionality breaks down' is a criterion choice, not a physical phase transition. Requiring 2-sigma or 3-sigma separation reduces T* by factors of 4 or 9; the alternative in SI Eq. 4 is still a 1-sigma overlap criterion and is lower. The paper should report T* for at least two separation thresholds and show how the Fig. 4 window depends on this choice, because the claim of biological robustness is sensitive to it.","section":"Eq. (8), Sec. III; SI, 'Alternative Definition for Crossover Condition'"},{"comment":"The dimensionless-to-real temperature conversion uses k∼(L/5)Y with Y≈10 GPa and L≈5 nm as fixed values. The gray bands in Figs. 2–4 propagate only the inter-protein variability in L and ε_s (or L_M), not the uncertainty in Y or in the effective spring stiffness. Since Θ in SI Eq. 3 is inversely proportional to Y and L^3, a factor of 2 uncertainty in Y shifts the 300 K line substantially. Please propagate these uncertainties or justify the fixed values with cited ranges; this affects whether the claimed (Δ,r) window survives.","section":"Sec. VII C and SI Eq. (3)"},{"comment":"The Discussion states that because training is at T=0, all finite-temperature effects are energetic fluctuations about the trained state, and that entropic mechanisms, prestress, or frustration are not included. This is an explicitly acknowledged limitation, and it is load-bearing for the extrapolation to real protein cooperativity (though not for the internal network result). The conclusions should state more cautiously that the biological relevance claim applies to energetic mechanisms only, and should identify a concrete test—e.g., training with prestress or a finite-temperature cost function—that could distinguish the two regimes.","section":"Sec. VI, Discussion"}],"minor_comments":[{"comment":"Typos and wording issues: 'Dimentional' (SI headings), 'T o' in Fig. S17 caption, 'strick' in the εs section, 'yeilding' in the same section. Please proofread.","section":"SI, various"},{"comment":"The gray dashed line and shaded band are described as '300 K' but they are estimates based on several model assumptions. Consider labeling them as '300 K estimate (median/quartiles)' and noting the conversion in every caption where they appear.","section":"Fig. 2D / Fig. 3 captions"},{"comment":"The symbol q in SI Eq. (3) is defined as ε for Θ* and L_M for ΘM, while the main text uses ε_s in Eq. (8). Use one consistent notation (e.g., ε_s always) to avoid confusion.","section":"SI Eq. (3) and Eq. (8)"},{"comment":"The definition of Ecoop is clear, but the sign convention for EA|B vs EB|A is not obvious from the text alone. A sentence explicitly stating that lower EA|B corresponds to facilitation would help.","section":"Sec. II, Eq. (4)"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely publishable after the authors address the sensitivity of the biological-temperature claim. The internal network results are solid, but the current presentation overstates the quantitative support for the 300 K window. I would encourage the editor to require a robustness analysis of the εs and 1-sigma choices rather than a full rewrite."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nHere's my take: this is a legitimately useful computational study. The new pieces are a clean linear-response treatment of thermal strain fluctuations in trained elastic networks, a crossover temperature T* defined by when unbound fluctuations reach the trained response, and a modified contrast function (the r term) that raises T* by stiffening the cooperative mode. The MD validation in the SI is real evidence: roughly 6% average error up to T* across networks trained at several Delta. The symmetric bidirectional training for cooperative binding is a natural extension of earlier unidirectional allostery work, but it works, and the paper shows it clearly raises the cooperativity parameter R.\n\nThe soft spots are all on the 'biologically relevant temperatures' claim. The conversion from simulation units to Kelvin leans on two choices: a source strain epsilon_s taken at the 83rd percentile of intra-protein atom-pair strains (optimized via a score function that involves a trade-off), and a 1-sigma separation as the crossover criterion. The 25th-75th percentile range for epsilon_s spans 0.09-0.31, so the gray band already includes a 4-5x range in the temperature conversion; the circled points in Fig. 4 apparently survive even the 0.09 end, which is good. But if you require 2-sigma separation for 'robust' function, T* drops by a factor of 4, and that would likely erase most of the Fig. 4 window. The raw per-pair median strain would be far lower than 0.19, though the authors' argument that such pairs are uninformative for a coarse-grained model is not unreasonable---it is a modeling choice, not an error. Also, the paper reports only successfully trained networks; no failure rate is given, which makes it hard to judge how common the useful cases are.\n\nNone of this undermines the internal mechanics: the linear-response formula, the MD validation, and the r-training protocol stand on their own. The limitation that training is at T=0 and energetic is stated openly in the Discussion, and I agree it is secondary given the proof-of-principle framing.\n\nI'd send this to referees. It's a solid paper for the mechanical-networks community, and the crossover framework will likely be reused. Ask the authors to add a robustness check on the sigma criterion and to report training success rates; the 300 K language should be toned down or made explicitly illustrative.\n\nVerdict: conditional, worth a serious referee.","headline":"Good thermal-fluctuation framework for trained networks, but the 300 K robustness claim depends on a lenient 1-sigma criterion and a strain calibration that is a judgment call.","tokens_in":22747,"tokens_out":5155,"would_cite":true,"duration_ms":86417,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Elastic networks trained for symmetric bidirectional allostery can exhibit cooperative binding that remains functional at biologically relevant temperatures.","keywords":["cooperative binding","allostery","elastic networks","thermal fluctuations","crossover temperature","coupled learning","mechanical training","protein models"],"falsifier":"Using elastic-network normal modes of real cooperative proteins (e.g., hemoglobin), compute T* from Eq. 6 with measured source strains; if the resulting T* values cluster below 300 K for proteins that are known to cooperate at that temperature, the energetic mechanism proposed here cannot account for protein robustness.","tokens_in":21731,"feed_emoji":"🧬","tokens_out":6469,"duration_ms":44140,"temperature":0.7,"pith_summary":"This paper shows that a disordered spring network can be trained, by applying allosteric training symmetrically at two sites, so that straining either site lowers the energy cost of straining the other—the signature of cooperative binding. The authors derive a crossover temperature T* at which thermal fluctuations at one site are as large as the signal from the other, and they show that T* is non-monotonic in the trained response amplitude and vanishes at the fully cooperative limit. They then modify the training cost function to selectively stiffen the low-frequency mode that carries the cooperative response, which raises T*. Using a Lindemann criterion for melting, they find a window of training parameters in which both T* and the melting temperature lie above 300 K, their proxy for biologically relevant conditions. The authors present the result as a proof of principle that thermal stability of cooperative function is achievable in model elastic networks.","feed_headline":"Trained spring networks can keep cooperative binding at 300 K","feed_subtitle":"Stiffening the functional mode during training pushes both crossover and melting temperatures above 300 K.","key_machinery":"The load-bearing object is the cooperative mode, the low-frequency normal mode onto which the symmetric strain response projects. Because the thermal strain variance at a site is a sum over modes of kBT/(mω²) times the mode's projection onto that site, a soft mode both enables large response and invites large thermal noise. The training modification is a generalized contrast function C̃ = (1−r)(EC−EF) − rEF, which for small r preferentially stiffens the bonds that store energy in the free state, raising the functional mode's frequency while preserving the trained response. This trade-off is what opens the window where both T* and TM exceed 300 K.","core_discovery":"Cooperative binding in a mechanical network can be trained as bidirectional allostery: the same allosteric response amplitude Δ is imposed from site A to B and from B to A. The training uses a symmetric version of the coupled-learning contrast rule, and the resulting function is carried by a single low-frequency vibrational mode. The paper derives the one-sigma thermal strain fluctuation at a target site from normal modes (Eq. 6) and defines the crossover temperature T* as the point where that fluctuation equals the trained mean response. A generalized contrast function with parameter r trades off a small amount of training accuracy for a higher free-state energy, which raises the frequency","pith_inferences":["If evolution can tune protein stiffness distributions the way this training does, the same trade-off between functional-mode frequency and overall softness may explain why some cooperative proteins are more thermally stable than others.","The crossover criterion based on 1-σ fluctuations is permissive; the SI's alternative that includes bound-state fluctuations gives lower T*, so real proteins may need even larger margins than the reported window.","Uniform stiffening of all springs also raises the dimensional T*, but the paper's dimensionless analysis implies that the non-trivial design rule is to concentrate stiffness in the functional mode—a testable principle for synthetic metamaterials.","The entropic-mechanism caveat suggests a concrete follow-up: train networks at finite temperature or with prestress, and compare the resulting T* to the T=0-trained case; if finite-temperature training yields lower T*, the energetic mechanism is not sufficient for naturally evolved proteins."],"forward_implications":["Cooperative binding can be trained as a symmetric allosteric task, and it produces higher cooperativity than unidirectional allostery training, with the cooperativity parameter R approaching 1 as Δ→1.","The crossover temperature is non-monotonic in the response amplitude: it rises with Δ for small Δ, then falls as the cooperative mode softens, vanishing at both Δ=0 and Δ=1.","A small penalty for free-state energy (r up to about 5×10⁻⁶) raises the dimensionless crossover temperature; larger r stiffens less important bonds and lowers Θ* while still raising the dimensional T*.","There exist trained networks for which both the functional crossover temperature and the melting temperature lie above the 300 K protein-equivalent threshold, so thermal robustness is achievable in principle.","The linear-response prediction of strain fluctuations, Eq. 6, matches molecular dynamics simulation quantitatively even beyond the linear regime and up to two orders of magnitude above T*."],"fun_headline_variants":["Bidirectional allostery keeps spring networks cooperative at 300 K","Bidirectional coupling trains heat-robust spring networks","Stiffen the functional mode to push spring networks past 300 K","Bidirectional allostery boosts heat limit of trained spring networks","Heat-proofing spring networks by symmetric allostery tuning"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"Because training is done at zero temperature, the cooperative mechanism is purely energetic; if real protein cooperativity involves significant entropic effects, prestress, or frustration, the thermal stability measured here may not transfer to proteins.","fun_headline_variants_meta":{"raw":{"variants":["Bidirectional allostery keeps spring networks cooperative at 300 K","Bidirectional coupling trains heat-robust spring networks","Stiffen the functional mode to push spring networks past 300 K","Bidirectional allostery boosts heat limit of trained spring networks","Heat-proofing spring networks by symmetric allostery tuning"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001404,"raw_usage":{"total_tokens":5448,"prompt_tokens":614,"completion_tokens":4834,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":358,"completion_tokens_details":{"reasoning_tokens":4748}},"tokens_in":358,"tokens_out":4834,"duration_ms":23886,"temperature":1.0,"reasoning_tokens":4748,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T15:12:50.734517+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Using elastic-network normal modes of real cooperative proteins (e.g., hemoglobin), compute T* from Eq. 6 with measured source strains; if the resulting T* values cluster below 300 K for proteins that are known to cooperate at that temperature, the energetic mechanism proposed here cannot account for protein robustness.","supporting_citations":[],"review_version":1}