Back

Robust error-minimization in the genetic code across physicochemical metrics and variant codes: a graph-theoretic analysis in GF(2)6

Clayworth, P.; Kornilov, S.

2026-04-29 genetics
10.64898/2026.04.25.720843 bioRxiv
Show abstract

The standard genetic code reduces the impact of point mutations, but the robust-ness of this property across physicochemical metrics, naturally variant codes, and codon-reassignment mechanisms remains incompletely quantified. We embed the 64 codons in GF(2)6, representing the hypercube Q6 as a coordinate-dependent subgraph of the encoding-independent single-nucleotide mutation graph H(3, 4), which supports continuous{rho} -interpolation between the two and enables joint analysis of physicochemical error minimization and codon-family topology. Under a block-preserving null (n =10,000), the standard code is significantly low-cost across four distinct amino-acid distance metrics (Grantham p = 0.006; Miyata p < 0.001; Woese polar requirement p = 0.003; Kyte-Doolittle hydropathy p = 0.001), addressing the concern that prior optimality results could be metric-specific; the signal strengthens monotonically as{rho} moves Q6 [-&gt;] H(3, 4). Across the 27 NCBI translation tables, near-optimality is broadly preserved: 11 of the 12 infor-mative-distance variants retain top-5% placement after BH-FDR correction (yeast mitochondrial is the sole marginal exception). Natural codon reassignments rarely break codon-family connectivity: under H(3, 4), only 6 of 28 observed events are topology-breaking versus 66% of 1,280 candidate moves (RR 0.32, permutation p [&le;] 10-4). This depletion is robust to alternative topology definitions, clade exclusions, and base-to-bit encodings, although the small breaker subset (4 of 6 from a single yeast-mitochondrial lineage; denominator effect in clade-exclusion robustness) is underpowered for strong cross-clade inference. Conditional-logit decomposition shows that topology avoidance and local physicochemical cost provide complementary, only weakly correlated signal (rs = 0.15); a heuristic tRNA-distance proxy does not improve fit, and several variant-code lineages show suggestive tRNA-gene enrichment for reassigned amino acids. Retrospective reanalysis of nine genome-recoding datasets is consistent with--but does not establish--a working hypothesis in which codon-family topology operates at a different biological layer from acute cellular fitness: Syn61 tolerated 18,218 boundary-crossing serine swaps as a class (genome-wide, not per-codon-position viability), yet the same move type is 3.1-fold depleted across natural code evolution. The contribution is the second axis: code evolution is jointly constrained by physicochemical smoothness and codon-family topological integrity, and these two constraints are partly independent. HighlightsO_LICodon-space geometry links genetic-code robustness and reassignment paths C_LIO_LIStandard and variant codes preserve broad physicochemical error minimization C_LIO_LIReassignments are depleted for codon-family topology-breaking moves C_LIO_LIConditional-logit models separate topology from physicochemical similarity C_LIO_LISynthetic recoding shows boundary conditions for natural-code constraints C_LI

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.