Sanskrit · Research

The Meter and the Molecule

Part Two: does the gene scan?

by Renay Oshop · August 2026 · Part Two of a series

Part One of this series ended with a dictionary and a dare. The dictionary maps every codon of the genetic code onto one of the eight gaṇas of Sanskrit prosody — exactly, as a projection, in two defensible polarities. The dare was to spend it: transcribe real genes into meter and ask, with honest statistics, whether anything in a living sequence scans. This essay does that. I will tell you the ending now, because the ending is the point: the gene hides no ślokas, one tantalizing meter turned out to be a mirage with a perfectly ordinary explanation, and the two findings that survive are the two the sober reader of Part One would have predicted — which is, in its own way, the most encouraging outcome this project could have had.

The corpus, and the paranoia

Two genes carry the analysis, both chosen for being small, iconic, and independently checkable. The first is HBB, human beta-globin — the protein of red blood cells, the gene of sickle-cell fame, and molecular biology's oldest textbook example: 1,608 bases on chromosome 11, three exons interrupted by two introns of 130 and 850 bases.² The second is INS, human insulin, whose 333-base coding sequence is nearly the mirror of beta-globin in composition — insulin's coding sequence runs 64.6% strong bases (G·C) to beta-globin's 56.1%.⁴

Because an analysis of this kind is only as good as its text, the sequences were treated the way a philologist treats a manuscript. Each was retrieved from two independent archives — NCBI RefSeq and Ensembl — and accepted only when the two witnesses agreed to the letter.²³ The beta-globin gene had to pass a further gate: its three exons, cut from the genomic text at Ensembl's annotated coordinates, had to reassemble letter-perfectly into the independently fetched mRNA; both introns had to begin with GT and end with AG, the nearly invariant splice signal of eukaryotic genes;⁵ and each coding sequence had to translate into its known protein, beta-globin's famous opening (the one whose sixth position, mutated, gives sickle-cell disease⁶) included. Every permutation test below is seeded and reproducible, and the complete code, sequences, and results are in the download box at the end. Statistics are only as honest as they are checkable.

The gene, recited

Transcription is mechanical, which is its virtue. Under the S/W polarity — strong pairs G·C as guru, weak pairs A·T as laghu, the bond-strength reading that Part One argued is closest in spirit to what guru means — each base becomes a syllable weight and each codon becomes a gaṇa. Beta-globin's coding sequence is 148 codons: 148 gaṇas, a poem of 444 syllables that runs 693 mātrās. Its first line, read straight off the DNA, is sa-ra-bha-ra-ja-ta… under one polarity and ra-ra-ja-sa-bha-na… under the other — Figure 1 shows the opening sixteen codons both ways, with the protein they spell.

Does anything scan?

Now the dare proper. The śloka — the eight-syllables-times-four meter of the epics, the most forgiving verse form in Sanskrit — was operationalized as its standard pathyā scheme: in every pāda, syllables five and six must run light-heavy; syllable seven must be heavy in the odd pādas and light in the even; syllables two and three may not both be light. Slide a 32-base window along a weight-string and every window either satisfies all four pādas or it does not. The beta-globin gene offers 1,577 such windows per polarity; each was tested, and then the whole procedure was repeated on ten thousand shuffles of the same sequence — same letters, same composition, order destroyed — to see what mere chance produces.

The result is as clean as a null result gets. Observed śloka-legal windows in the actual gene: zero, in both polarities, and likewise zero in both coding sequences. The shuffled gene also produces zero about 95% of the time (Figure 2); chance expects 0.06 legal windows per pass. The gene is not avoiding the śloka — it simply has no reason to contain one, and it doesn't. Whatever the genome is doing, it is not hiding epic verse at the base-pair level, and it would have been a red flag for this whole enterprise if it were.

The named meters of the classical catalog tell the same story with a better twist. Searching the gene for the exact binary pattern of each samavṛtta meter from Part One's catalog, most show zero or one hit — and one hit is exactly what chance predicts. An Upendravajrā does occur once in the gene under S/W; chance expected 0.51 of one. Finding a single eleven-syllable meter in sixteen hundred bases is precisely as remarkable as naming an eleven-coin-flip pattern in advance and then seeing it once in sixteen hundred flips: not at all.

One meter, though, flickered. Sragviṇī — ra ra ra ra, the garland-weaver, pattern GLGGLGGLGGLG — appears three times in the gene under R/Y, where chance expects 0.27. Taken alone that is p ≈ 0.009, and for an afternoon it was exciting. Then the two standard sobriety checks. First, this search examined twenty-one meters, in two polarities, in two sequence classes — a family of eighty-four tests, and a p of 0.009 corrected for eighty-four opportunities is p ≈ 0.74: unremarkable. Second, and more instructive: where are the three hits? All in intron 2, clustered within ninety bases of each other, inside A·T-rich stretches like GCAATAATGATACAA — the repetitive, low-complexity sequence with which introns are littered. A purine-pyrimidine pattern with period three, repeated, is just what such repeat-rich DNA produces by the yard. The garland dissolves on inspection. I keep it in the essay because this is exactly how pattern-hunting in the genome goes wrong, one flicker at a time, and the discipline of watching one die is worth more than a dozen abstract warnings about multiple testing.⁸

The tempo of the gene

So the gene does not scan as verse. But mātrā meters, Part One noted, fix something subtler than a pattern: they fix a tempo, a duration per stretch of syllables. And here the dictionary earns its keep, because under the S/W polarity the mātrā count of a sequence is not an occult quantity at all. A guru is two beats, a laghu one, so mātrās-per-base is exactly one plus the fraction of strong bases — the prosodic reading is the G·C content, wearing different clothes. Whatever is true of GC content in genomes becomes, through the dictionary, a statement about metrical tempo.

And something is true. Beta-globin's exons run 1.51 mātrās per base; its introns run 1.33; ten thousand label permutations put the difference far beyond chance (p < 0.0001). Figure 3 draws the gene in true proportion as a tempo map: the passages the cell translates are held heavy and slow, the interludes it splices away run light and quick — with the long second intron, at 1.31, the lightest stretch of all. Under R/Y the tempo is flat (1.48 vs 1.44, p = 0.08), which is its own quiet confirmation: purine fractions hover near one-half on both strands and both region types, an echo of Chargaff's second parity rule,⁷ so a duration signal appears only in the polarity where real compositional structure exists to carry it.

Let me be precise about what this is and is not. The GC-richness of coding sequence relative to intronic and intergenic DNA is established genomics, part of the literature on isochores and codon composition;⁹ the meter lens discovers nothing here that a genome browser doesn't already know. What the lens does is translate it, and the translation is apt in a way that is hard to shake off: in the S/W polarity, guru literally means "more strongly bound, longer to melt apart" — thermodynamic weight — and it is the protein-coding passages, the functional heart of the gene, that the double helix holds heaviest. The tradition's intuition that weight marks significance maps, in this one narrow and checkable sense, onto the molecule.

The anceps position, measured

Part One offered one structural analogy with a promissory note attached: the pāda relaxes its final syllable (anceps), the codon relaxes its third position (wobble), and both codes concentrate tolerance at the boundary of the unit. The corpus lets us pay that note in numbers. Take every codon of beta-globin, consider every synonymous alternative — every other codon spelling the same amino acid — and ask: if the cell had chosen that spelling instead, would the syllable's weight change? At codon position two, essentially never: 0.2% of synonymous choices flip its S/W weight. At position one, 9.3%. At position three, 49.5% — half of all synonymous freedom is freedom to reweight the codon's last syllable (Figure 5; the R/Y numbers tell the same story at 5.0%, 2.7%, 34.4%). The codon really is built like a pāda: its interior weight is grammar, its final weight is licence.

The same structure explains the one place the gaṇa statistics do deviate from chance. Against a composition-matched null — random weight-strings with the gene's own guru fraction — beta-globin's gaṇa spectrum is skewed (χ² = 55.4, p = 0.0001 under S/W; Figure 4): a spectacular excess of ra — fifty occurrences against twenty expected — with ma, ya, and ja starved to pay for it, because codon positions carry systematically different weights and the reading frame stamps that period-three signature onto the triplets. (ra is G·L·G — heavy, light, heavy — the natural gait of codons whose first and third letters run strong while the middle runs weak.) Insulin shows the same S/W skew (χ² = 45.2) — yet under R/Y its spectrum is indistinguishable from chance (χ² = 2.8, p = 0.90). The deviations, where they exist, are the fingerprint of the reading frame and codon usage — real, explicable structure, not hidden authorship.

What survives

The scoreboard, then, dated and honest. Dead: hidden ślokas, and any version of the idea that genes conceal classical verse — the nulls were clean in every sequence and both polarities. Instructively dead: the Sragviṇī flicker, three garlands that were really microsatellite lint in an intron, p = 0.74 after correction. Alive: the tempo result — the dictionary faithfully translates the genome's compositional architecture into duration, and the translation runs in the semantically right direction, heavy on the functional passages; and the anceps result — the third position's metrical freedom is now a measured 49.5%, not a metaphor. Alive, too, is the method itself: seeded nulls, corrected p-values, and sequences vetted like manuscripts turn out to make "Sanskrit meets molecular biology" a workable empirical genre rather than a mood.

Sequences were accepted only after exact agreement between NCBI and Ensembl copies, splice-reassembly of the beta-globin gene, GT–AG boundary checks, and translation to the known proteins. All permutation tests are seeded (seed 20260812, 10,000 permutations) and every number in this essay is generated by the downloadable code; a consistency script diffed the prose against results.json before publication.