{"id":"2f279687-7f10-48e5-95e3-5ef9d69532e8","arxiv_id":"2502.03851","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":0.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A review of how local energetic frustration in proteins contributes to folding, function, and disease, with speculative extensions to gene networks and the brain.","lead":"This paper reviews the concept of local frustration in proteins, where some internal interactions are energetically suboptimal and this imperfection enables function. It argues that tracking frustration patterns helps explain physiology, disease mutations, and potentially drug design.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central claim depends on the frustration index being robust to modeling choices; the same site can flip from highly to minimally frustrated when ions or partners are included, so the pathology correlations built on these labels are not yet secure.","rationale":"The reader's conditional verdict is well-calibrated. The review is not purely speculative: it cites experimental checks (NMR order parameters in thrombin, smFRET in NF-kappaB, ddPCA stability correlations, AlphaFold2 re-prediction of RfaH and AdK conformations), which give real support for specific applications. My concern is narrower: the general label \"local frustration\" is assigned by a computational pipeline whose output depends on choices that the paper itself shows are consequential (Sections 9.3 and 9.4). If the labeling is not robust to force field and structural context, then the broad claims - that frustration is an essential part of protein inner workings and that frustration changes are linked to pathologies - are not yet established, even if each case study is internally coherent. A focused robustness test would settle this and would be feasible with the public Frustratometer tools. This does not change the reader's conditional verdict, but it sharpens the condition: the claims should be presented as dependent on the chosen frustration measure until the robustness check is done.","tokens_in":61480,"tokens_out":10607,"duration_ms":125105,"concrete_test":"Run a systematic robustness analysis on the datasets that ground the central claim (catalytic/binding-site enrichments in Sections 4 and 10, disease-variant correlations in Section 11.11, and the representative pathology cases in Sections 11.1-11.10): (i) sweep the frustration-index thresholds across a plausible range (e.g., highly frustrated cutoff from -1.5 to -0.5, minimally frustrated cutoff from 0.5 to 1.5) and (ii) recompute labels with the all-atom Rosetta Frustratometer including crystallographic heteroatoms and binding partners where known, as illustrated for CaM in Section 9.4.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Section 3, \"mutational\" and \"configurational\" frustration are defined by comparing native contact energies to decoy ensembles under the AWSEM potential, with fixed cutoffs (the review later uses FI ≤ -1 for highly frustrated and FI > 0.78 for minimally frustrated, Fig. 13). The central claim that frustration is an essential, functionally selected property treats these labels as intrinsic biophysical observables. But the labels are not stable under modeling choices. Section 9.3 shows the CaM Ca2+-binding loop is \"highly frustrated\" in the coarse-grained Frustratometer only because calcium is excluded; with the all-atom version including the ion, those interactions are minimally frustrated (Section 9.4). More broadly, Section 4 reports that interface sites become \"highly frustrated\" when binding partners are removed from the calculation. Thus the index is partly reporting omitted structural context (ions, ligands, partners), not necessarily an evolved internal conflict between folding and function. The large-scale disease-variant correlation (Section 11.11, Kumar et al.) and every pathology vignette in Sections 11-12 use this same index, so the abstract's inference that frustration alterations \"can be linked to distinct pathologies\" inherits the ambiguity. The issue is not whether the method is useful: it has real correlations with NMR, smFRET, and ddPCA data in specific cases. The issue is that its categorical labels must be robust to force-field and context changes before the general physiological and medical claims are secure.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This review by Parra et al. surveys the energy-landscape concept of local frustration and its applications to protein folding, function, evolution, pathology, and higher-order biological networks. It introduces the two frustration indices (mutational and configurational), summarizes evidence from the Frustratometer and AWSEM, and walks through a series of case studies in which frustration patterns correlate with binding, catalysis, allostery, chaperone action, and disease mutations. The final sections extrapolate the concept to gene regulatory networks and brain function. The abstract's central claim is that local frustration is an evolutionarily selected, functionally essential property and that frustration alterations can be linked to pathologies.","tokens_in":61761,"tokens_out":4941,"duration_ms":53578,"significance":"The review is a timely, broad synthesis from the group that developed the frustration analysis tools, and it collects a large body of computational and experimental evidence in one place. Its strengths include the clear presentation of the physical origins of frustration, the inclusion of independent experimental validations (smFRET, NMR relaxation, ddPCA data), and an unusually candid statement of some limitations (e.g., the coarse-grained Frustratometer's blindness to heteroatoms in Section 9.3). If the central claim holds, the frustration index provides a practical descriptor that localizes functional regions, interprets disease mutations, and can be integrated into drug design. However, the manuscript's strongest predictive claims depend on the robustness of the categorical frustration labels and on the validity of a near-equilibrium landscape formalism for networks; both remain only partially established.","major_comments":[{"comment":"The operational definitions of mutational and configurational frustration in Section 3 compare native contacts to decoy ensembles under the AWSEM potential, and the categorical assignment later is made via thresholds (highly frustrated for FI ≤ -1 and minimally frustrated for FI > 0.78 in Fig. 13). Section 9.3 shows that the CaM Ca2+-binding loop appears highly frustrated in the coarse-grained Frustratometer only because Ca2+ is excluded, and the all-atom version including the ion reclassifies those interactions as minimally frustrated. Section 4 likewise reports that interface sites become highly frustrated when binding partners are removed. Because the large-scale disease-variant analysis in Section 11.11 and most pathology vignettes use this same index, the abstract's inference that frustration alterations 'can be linked to distinct pathologies' inherits this context dependence. I ask the authors to add an explicit robustness discussion and, where feasible, a quantitative comparison of coarse-grained versus all-atom frustration classifications for the disease-associated sites discussed in Sections 11-12.","section":"§3, §9.3, §11.11"},{"comment":"The extension to gene regulatory networks and to brain circuits is framed in quantitative terms ('minimally frustrated' gene networks; epilepsy as 'too little frustration'), but Section 14 itself states that 'it is not yet established how best to quantify frustration for complex many body systems that do not possess quasi-stable states' and cites non-gradient gauge-field descriptions for far-from-equilibrium systems. Since the abstract promises extensions to gene regulatory networks and neural networks, the manuscript should either introduce the appropriate non-equilibrium formalism or explicitly relabel Sections 13-14 as qualitative analogies, with the limitations stated in the abstract as well as in the text.","section":"§13, §14"},{"comment":"The large-scale correlation between single-nucleotide variants and frustration changes is described only as 'more substantial changes' and 'greater frustration changes', without effect sizes, confidence intervals, or a comparison with established variant-prediction baselines. Given that this is the main quantitative support for the claim that frustration alterations can be linked to pathologies, the review should report the actual statistics from Kumar et al. (correlation coefficients, AUC, or equivalent) or state explicitly that such measures were not reported in the source study.","section":"§11.11"},{"comment":"The thrombin example is presented as a key experimental validation: residual frustration correlates with the S2AMD order parameters 'strikingly'. No correlation coefficient, p-value, or effect-size measure is given for this comparison, although the corresponding S2ns comparison is explicitly non-correlating. Adding the quantitative measure would substantially strengthen the claim that frustration patterns report long-time dynamical regions central to function.","section":"§5.2"}],"minor_comments":[{"comment":"There are inconsistent name spellings: 'Tsai1' should be 'Tsa1', and 'Steltz' should be 'Stelzl' in the text describing the DsbD work.","section":"§7"},{"comment":"The term 'Sabercoviruses' appears repeatedly and should be corrected to 'Sarbecoviruses'.","section":"§9.2"},{"comment":"The text refers to 'Figure 2B' for the comparison of residual frustration with S2AMD order parameters; this should be 'Figure 3B', which is the panel that actually shows the comparison.","section":"§5.2"},{"comment":"The statement that Cys145 establishes 18 minimally frustrated interactions in the new-inactive conformation versus 8 in the active one lacks context; it should specify whether these are per-residue contact counts and whether the difference is statistically meaningful.","section":"§11.10"},{"comment":"Several case studies perform frustration calculations in the absence of cofactors or ligands (PLP in AADC, Ca2+ in cTnC, RNA in MS2); as the authors correctly flag for CaM in Section 9.3, this context dependence should be stated in each relevant figure caption or case summary so readers do not interpret the labels as intrinsic to the isolated protein chain.","section":"§11.4, §11.10, §12"}],"recommendation":"major_revision","confidential_remarks":"This is a broad review from the group that developed the frustration index, and the citation pattern is correspondingly self-referential. That is largely appropriate for a review, but the editor may wish to check that the quantitative claims in Sections 5.2 and 11.11 are accurately reported from the cited primary sources, since the manuscript itself does not provide the underlying statistics. The main concern is whether the categorical frustration labels are stable enough to support the pathology-link claims; a robustness analysis or a more cautious framing would resolve this."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things. First, this is a review, not a new-results paper: no new derivations, data, or predictions. Second, it is the most complete and readable survey of local frustration in proteins that the group has produced, and it makes a surprisingly strong case that frustration is a useful organizing principle for interpreting protein function and disease mutations.\n\nWhat it does well: it systematically explains the physical origins of frustration, defines mutational and configurational indices clearly, and then walks through concrete systems—NF-κB, thrombin, chaperones, disulfide bonds, metamorphic proteins—where frustration patterns correlate with independent experimental observables like NMR relaxation, smFRET, and ddPCA. The sections on chaperone-client recognition and viral assembly are genuinely informative and not just hand-waving. The writing is clear enough that a graduate student could follow the logic.\n\nThe soft spots are real but proportionally modest. Novelty is zero, and the self-citation is heavy—that is expected in a group’s own review, but it matters for who should referee it. The stress-test concern about the model-dependence of the frustration index is valid: the coarse-grained Frustratometer excludes ions and partners, so a site can flip from highly frustrated to minimally frustrated when calcium or a binding partner is included. The paper does acknowledge this in Section 9.3 for calmodulin, which is honest, but the abstract’s general claim that frustration alterations 'can be linked to distinct pathologies' inherits that ambiguity. The extensions to gene regulatory networks and psychiatry are speculative, and the paper itself admits that much remains unclear. Those sections are weaker, but they are clearly framed as exploratory.\n\nThe central argument—that local frustration is a selected, functionally important feature of proteins—holds up well and has support from experiments independent of the group’s own models. The model-dependence is a caution, not a fatal flaw.\n\nWho is this for? Anyone working on protein biophysics, protein evolution, or the interpretation of disease variants. I would cite it as a comprehensive entry point. This deserves a serious referee—not for novelty, but for accuracy and for tempering universal claims. A good revision would add an explicit statement about the index’s context-dependence wherever pathology claims are made.","headline":"A comprehensive, clearly written review of the frustration framework from the group that built it; the model-dependence of the index is real and acknowledged, but the central functional claims hold up.","tokens_in":62290,"tokens_out":1763,"would_cite":true,"duration_ms":19623,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Folded proteins keep a small share of suboptimal, 'frustrated' contacts, and this review argues those contacts are essential for binding, catalysis, and allostery while mutations that shift them are linked to disease.","keywords":["protein frustration","energy landscapes","minimal frustration principle","protein folding","protein-protein recognition","allostery","disease mutations","frustration index"],"falsifier":"A direct test: for a panel of proteins, measure local stability experimentally at residues the index labels highly frustrated; if those residues are not systematically less protected and more mobile than neutral-class residues, the index is not reporting the physical quantity the review claims. Separately, a genome-scale collection of disease-associated and benign missense variants should show larger frustration shifts in the disease set after controlling for burial and sequence conservation; failure of that enrichment would falsify the pathology link.","tokens_in":1715,"feed_emoji":"🧬","tokens_out":2237,"duration_ms":97057,"temperature":0.7,"pith_summary":"The paper tries to establish that local frustration—the persistent inability of a folded protein to satisfy all its energetic interactions at once—is not a defect but a design feature. It argues that a quantifiable frustration index, computed by comparing each native contact against a distribution of energies of randomly mutated or reshuffled alternatives, reveals that roughly 10% of interactions are highly frustrated, clustered at the surfaces and functional regions where proteins bind partners, catalyze reactions, and switch conformations. The review marshals evidence that these frustrated patches are evolutionarily conserved and that disease-causing mutations often change frustration patterns, making frustration analysis a plausible tool for finding functional sites, interpreting mutations, and guiding drug design. It closes by extending the same energy-landscape reasoning to gene regulatory networks and brain circuits, where minimal frustration would keep attractor states stable and too much or too little frustration would produce pathology.","feed_headline":"Local frustration drives protein function and disease","feed_subtitle":"~10% of poorly optimized contacts power binding, catalysis, and allostery; mutations that shift them cause disease.","key_machinery":"The carrier of the argument is the local frustration index. For each native contact in a protein structure, the energy of the native sequence is compared with the energy distribution of a decoy ensemble: either 399 random amino-acid substitutions at the contact, giving the mutational frustration index, or structurally reshuffled local configurations, giving the configurational frustration index, evaluated with the AWSEM coarse-grained force field. A contact whose native energy is much more favorable than the decoys is minimally frustrated; one near the decoy mean is neutral; one less favorable than most decoys is highly frustrated. The paper maps these classes onto structures and shows that highly frustrated contacts cluster at functional sites, so the index turns a static structure into a prediction of where dynamics, binding, and conformational switching will occur.","core_discovery":"The central claim is that a folded protein's energy landscape is a funnel that is only globally smooth: locally it retains frustrated contacts that are energetically suboptimal relative to random sequence or structural alternatives. Using the AWSEM coarse-grained energy function, the paper's authors and collaborators compute two indices—mutational frustration, which compares the native energy of a contact with the energies of all 399 possible amino-acid substitutions, and configurational frustration, which compares it with reshuffled local structures—and classify contacts as minimally frustrated, neutral, or highly frustrated. The paper states that about 40% of contacts are minimally frustrated and buried, 50% are neutral, and 10% are highly frustrated and clustered, often on surfaces; these frustrated patches coincide with protein-protein recognition sites, catalytic residues, allosterically mobile loops, and chaperone-binding regions. It further claims that disease-associated mutations tend to alter these patterns, either increasing frustration at core or interface positions or relieving it in ways that shift conformational equilibria, so that local frustration patterns are a bridge between energy landscape physics and molecular medicine.","pith_inferences":["If the disease-link holds generally, frustration-change scores could be added to variant interpretation pipelines as a continuous biophysical feature, alongside conservation and population frequency, for any gene with a structure or model available.","Because highly frustrated residues tend to be dynamic yet specific, they may be ideal handles for designing allosteric modulators: molecules that bind a frustrated surface patch could bias an ensemble without competing at the active site.","The review's leap from proteins to brains is the most speculative step; a concrete test would be whether frustration indices computed from connectome or EEG data show the predicted U-shape, with healthy brains intermediate between over-synchronized and chaotic regimes."],"forward_implications":["On structures alone, frustration maps can flag protein surfaces that bind partners, catalytic residues, and loops that move during allostery.","A mutation's pathogenicity can be assessed by whether it shifts frustration patterns at functional sites, not just by whether it reduces stability.","Drug candidates can be ranked by whether they form minimally frustrated interfaces with their target; strong inhibitors make many minimally frustrated contacts with the binding pocket.","Chaperones such as Spy and GroEL appear to recognize frustrated patches, which would explain their ability to bind diverse clients without a shared sequence motif.","The same minimal-frustration logic transfers to gene regulatory networks and neural circuits, where too little frustration can yield hypersynchronous states such as seizures and too much can yield chaotic, hallucinatory states."],"supporting_citations":[{"why":"Foundational energy-landscape theory showing natural proteins obey a principle of minimal frustration, which the review's measure interprets.","marker":"Bryngelson and Wolynes 1987"},{"why":"Introduces the local frustration calculation and reports the 40/50/10 distribution and the enrichment of frustration at binding interfaces.","marker":"Ferreiro et al. 2007"},{"why":"The earlier review that consolidated frustration as a functional feature of biomolecules and frames the current survey.","marker":"Ferreiro, Komives, and Wolynes 2014"},{"why":"Shows that regions changing structure in allosteric proteins are enriched in highly frustrated interactions.","marker":"Ferreiro et al. 2011"},{"why":"Large-scale survey showing catalytic sites and cofactor-binding sites are enriched in highly frustrated, conserved residues.","marker":"Freiberger et al. 2019"},{"why":"Introduces the all-atom frustration index and links minimally frustrated drug-target contacts to binding affinity.","marker":"M. Chen et al. 2020"},{"why":"Provides the updated frustratometeR tool used to track frustration changes along molecular dynamics trajectories and under mutations.","marker":"Rausch et al. 2021"},{"why":"Genome-scale evidence that disease-associated single-nucleotide variants change local frustration more than benign variants, with oncogenes and tumor suppressors showing opposite patterns.","marker":"Kumar, Clarke, and Gerstein 2016"},{"why":"Transfers the minimal-frustration attractor picture to gene regulatory networks and cell-type counting.","marker":"Sasai and Wolynes 2003"},{"why":"Applies frustration to viral capsid assembly, proposing programmed breaking of packaging-signal contacts as a regulatory principle.","marker":"Twarock, Towers, and Stockley 2024"}],"fun_headline_variants":["Frustrated protein contacts: 10% power binding, catalysis, disease","Local frustration: the hidden driver of protein function and disease","Protein frustration maps reveal disease mutation hotspots","10% of protein contacts are frustrated—and that's key"],"cache_read_input_tokens":64384,"weakest_assumption_plain":"The load-bearing premise is that the frustration index computed by comparing native contacts against randomly mutated or reshuffled decoy structures with a single coarse-grained energy function faithfully captures the thermodynamic frustration that operates inside living cells.","fun_headline_variants_meta":{"raw":{"variants":["Frustrated protein contacts: 10% power binding, catalysis, disease","Local frustration: the hidden driver of protein function and disease","Protein frustration maps reveal disease mutation hotspots","10% of protein contacts are frustrated—and that's key"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000174,"raw_usage":{"total_tokens":1284,"prompt_tokens":948,"completion_tokens":336,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":564,"completion_tokens_details":{"reasoning_tokens":268}},"tokens_in":564,"tokens_out":336,"duration_ms":4527,"temperature":1.0,"reasoning_tokens":268,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T00:29:27.861867+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct test: for a panel of proteins, measure local stability experimentally at residues the index labels highly frustrated; if those residues are not systematically less protected and more mobile than neutral-class residues, the index is not reporting the physical quantity the review claims. Separately, a genome-scale collection of disease-associated and benign missense variants should show larger frustration shifts in the disease set after controlling for burial and sequence conservation; failure of that enrichment would falsify the pathology link.","supporting_citations":[],"review_version":1}