Human mutation field reveals an equilibrium-like structure with irreversible circulation
Isabella Caranzano, Daniel Maria Busiello, Stefano Priorelli, Amos Maritan, Piero Fariselli
q-bio.GN · cond-mat.stat-mech · cs.AI
Quick verdict
Hodge decomposition of human mutation biases: dominant equilibrium gradient plus reproducible CpG-enriched curl; convincing, minor data-provenance caveats.
Summary
The paper treats seven-base context-dependent human germline mutation probabilities (Aggarwala & Voight) as an antisymmetric log-ratio field on directed substitution edges and asks whether it reduces to an equilibrium-like energy landscape or retains non-equilibrium circulation. Hodge projection of each four-node context graph shows a dominant gradient component (mean curl fraction 0.11) and an empirical reversible ceiling of r≈0.96. A Siamese network trained to predict only energy differences reaches r≈0.92–0.93 (Table I: 0.924 at r_int=4, 0.934 at r_int=7) versus 0.948±0.002 for an unconstrained MLP reference. Although trained solely on mutation probabilities, aggregated energies of the learned landscape correlate with GRCh38 word frequencies (r≈0.999 at ℓ=1, 0.889 at ℓ=4) and approximately satisfy Chargaff's second parity rule. Non-zero Kolmogorov cycle affinities (median |A|=0.451; ICC=0.91 across African, Asian and European populations) and small positive entropy production indicate weak, reproducible irreversibility, concentrated in CpG states and CpG-transition edges (top-10% odds ratios 16.1 and 28.7). Main limitations: probabilities derive from polymorphism data, so selection, ascertainment and GC-biased gene conversion may contaminate the inferred circulation; cycle-level tests are non-independent and only descriptive; composition recovery degrades with word length; and no external validation on de novo mutation data is provided.
Review complete
Assessment at a glance
3
risk flags identified
● high 0
● medium 2
● low 1
5 strengths · 4 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- The human seven-base mutation log-ratio field is gradient-dominated (mean curl fraction 0.11), and a Siamese energy model predicts it at r≈0.92-0.94, approaching the empirical Hodge ceiling of ≈0.96.
- Energies trained only on mutation directionality recover short-word genomic composition (r≈0.999 at ℓ=1 to 0.889 at ℓ=4) and approximately satisfy Chargaff's second parity rule.
- A small but reproducible irreversible circulation violates the Kolmogorov cycle condition, with ICC=0.91 across African, Asian and European populations and median |A|=0.451.
- The strongest cycle affinities are enriched in CpG-linked states and CpG-transition edges (top-10% odds ratios ≈16.1 and ≈28.7), consistent with methylcytosine deamination.
Strengths
Rigorous geometric framing with a data-derived ceiling
The Hodge projection (Eqs. 10-13) yields an empirical reversible ceiling r≈0.96, so the Siamese model's gap to the unconstrained MLP (0.92-0.93 vs 0.948, Table I) is attributable to genuine curl rather than architectural failure - a logically sound triangulation rarely seen in this genre.
Multiple independent consistency checks
Approximate stationarity via signed master-equation balance (Fig. 2b), composition recovery without training on genomic word frequencies (Fig. 3, r≈0.999 at ℓ=1 to 0.889 at ℓ=4), Chargaff second-parity agreement, and a genuine negative control grouping cycles by excluded base (Fig. 6d, indistinguishable |A| distributions).
Cross-population replication of the nonequilibrium signal
Biologically interpretable residual with effect sizes
Careful conceptual hedging
Weaknesses
Provenance confounding of the 'mutation field'
Cycle-level non-independence inflates enrichment statistics
Composition recovery partially anticipated and weakens with word length
Applicability claims not demonstrated
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper (arXiv:2609.07500, q-bio.GN/cond-mat.stat-mech/cs.AI) poses a statistical-physics question to genome biology: do context-dependent human mutation probabilities reduce to an equilibrium-like energy landscape, or do they retain irreducible non-equilibrium circulation? For each fixed six-base flank k, the authors define an antisymmetric log-ratio field φ_k(a,b) = log[P(ka→kb)/P(kb→ka)] on the 12 directed substitution edges of the four central bases (Section I, Eq. 1). Three central claims: (1) the field is gradient-dominated (mean Hodge curl fraction 0.11; empirical reversible ceiling r≈0.96) and approximately stationary under the master equation (Fig. 2b); (2) an energy landscape trained only on mutation directionality - via a Siamese network predicting exclusively energy differences - reaches r≈0.92-0.93 (Table I: 0.924 at r_int=4; 0.934 at r_int=7) versus 0.948±0.002 for an unconstrained MLP, and its Boltzmann-like word proxies recover GRCh38 short-word frequencies (r≈0.999 at ℓ=1 declining to 0.889 at ℓ=4) and Chargaff's second parity rule; (3) a small but reproducible irreversible component persists, with median |A|=0.451 cycle affinities, ICC=0.91 across African/Asian/European populations, positive entropy production, and strong CpG enrichment (top-10% odds ratios ≈16.1 for CpG state, ≈28.7 for CpG-transition edge), consistent with methylcytosine deamination.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 4 weaknesses, analyzed in detail
- 3 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is the 'human mutation field'?
For every seven-base sequence context, the paper computes log(P(forward substitution)/P(reverse substitution)) for each possible nucleotide change. This antisymmetric set of numbers over all directed substitution edges behaves like a field: its gradient part corresponds to an effective energy landscape, and its curl part measures irreversible, non-equilibrium circulation around closed mutation cycles.
Is human mutation at equilibrium (detailed balance)?
Mostly, but not exactly. About 89% of the field's squared magnitude is gradient-like (mean curl fraction 0.11) and an energy-only neural model predicts it with correlation around 0.93 against a ceiling of 0.96. Yet cycle affinities are measurably nonzero (median |A| = 0.451), reproducible across three populations (ICC = 0.91), and concentrated in CpG-related transitions, consistent with methylcytosine deamination.
How does the paper connect mutation bias to genome composition?
A Siamese network trained only on mutation log-ratios learns an energy for each seven-base context. Boltzmann-like aggregates of these energies correlate with observed GRCh38 word frequencies (r≈0.999 for single bases, 0.889 for 4-mers) and approximately satisfy Chargaff's second parity rule, even though the model never sees genomic word counts during training.
What are the main caveats of this study?
The mutation probabilities come from polymorphism data, so selection, ascertainment bias and GC-biased gene conversion may contaminate the inferred circulation; the cycle-level statistical tests are not fully independent; composition recovery weakens as word length grows; and no external de novo mutation dataset is used for validation.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figure-internal plotted values (box-plot spreads in Fig. 2a, entropy-production magnitudes in Fig. 5b, the top-20 cycle identities in Fig. 7) cannot be verified from the extracted text; only captions were available.
- Appendix A content appears in the extraction only as the Fig. 7 caption; no numeric listing of the top-ranked cycles was extractable.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.