From the point variant to structural RNA
QMetrika's thermodynamic engine was born for point variants: the nearest-neighbor stacking energy profile of the mRNA. There we showed that the synonymous pathogenicity signal lives in local stacking, not in secondary structure. That distinction is exactly the hinge toward repeat-expansion diseases: in a point variant, perturbing the global structure is noise; in an expansion, the periodic structure of the repeat is the relevant entity. The same local physics, in a new regime.
Counting repeats isn't enough
Myotonic dystrophy (DM1), Huntington's and the ataxia SCA1 share one clinical fact: age of onset correlates with repeat length, but weakly. The real modifier is interruptions — variant units embedded in the tract (CCG in DM1's CTG, the CAA in Huntington's CAG, CAT in SCA1). They reduce somatic instability and delay onset. The problem: the clinical observable is still a scalar, a count, blind to the pattern.
The observable: loss of periodicity
A pure repeat produces a periodic stacking profile. An interruption breaks that pattern locally. energoRNA quantifies it with a single number — the loss of autocorrelation (at the motif's period) of the Turner profile:
# periodicity_loss = 1 − ACF_period(Φ)
pip install energorna
energorna score --repeat CTG --n 40
It auto-calibrates the period per motif (3 for trinucleotides, 4 for DM2's CCTG tetranucleotide) and, unlike counting, it is sensitive to the type of interruption (CCG destabilizes the pattern far more than CTC) and to its position in the tract.
It beats counting, across four diseases
In DM1, the loss of periodicity predicts the onset residual far better than counting interruptions:
In Huntington's, the three canonical architectures rank by loss of periodicity exactly as by their published onset. In SCA1, with real, open, individual data (Menon 2013, n=35), the uninterrupted stretch — the observable's axis — predicts onset far above total size. And in Fragile X the same observable ranks the AGG architectures by their published expansion risk — a new motif, a new interruption and a new endpoint:
It's not how many repeats there are, it's the break in the periodic pattern of the energy field. The same physical observable, in DM1, Huntington's, SCA1 and Fragile X.
Isn't this just re-deriving the clinical metric?
The obvious critique: in SCA1/HD/Fragile X the established predictor is already the longest uninterrupted stretch. The answer, quantified: the loss of periodicity contains that axis but adds two dimensions a length scalar cannot represent — the type of interruption (at equal uninterrupted stretch, CCG/CTC/GGC give values 0.02–0.26) and its position (an identical block at 5' vs 3' gives the same stretch, but the 3'-weighted observable separates them 15×). And in real DM1, over the onset residual that length does not explain, the clinical scalar gives r=+0.09 while type (+0.81) and position (+0.84) capture it: energoRNA does not compete with length, it captures the interruption modifier that length is blind to.
Correlate, not cause — and ready to use
The loss of periodicity is a biologically motivated correlate — the periodicity of stacking is a proxy for the contiguous slipped-strand substrate on which the repair machinery (MMR) drives somatic expansion — not a demonstrated cause. It generates falsifiable predictions (correlation with measured somatic instability, UV melting, MSH3 binding). And because it's a physical scalar with no deep network, it's ready to use: numpy-only, no GPU, auditable.