PicoFold — predicting protein secondary structure from DNA

DRAFT EN — pending review by Pavel Pivovarov & A. Kushelev

What this is

PicoFold is a scientific web tool that predicts protein secondary structure from a DNA sequence (rather than from an amino-acid sequence, as AlphaFold or PSIPRED do). The algorithm is grounded in a ring-polyhedral model of amino acids (A. Kushelev, Nanoworld Laboratory) and follows a forward-kinematics principle: codon → helix type and angles → atomic coordinates.

What has been validated

• 15 proteins, 2,814 residues: Z-score 3.10 (p < 0.01) — a statistically significant correlation between the helix type and the third codon nucleotide.
• 113 proteins, 21,797 residues (DSSP):
  • Q3 = 0.377 (refined), random baseline = 0.347, gain +3.1 pp.
  • Helix signal: G-ending 39.5% H vs C-ending 30.1% H.
  • β-anti-correlation: A-ending = 18.8% strand (minimum).
  • Chi-square: 6 of 18 amino-acid families with p < 0.05 (Ser, Gly, Glu, Ile, Asp, Tyr).
  • Top proteins: Tropomyosin α (Q3 = 0.821), Liprin-α2 (0.726), Vimentin (0.703).

What this does not do

PicoFold does not predict the tertiary 3D fold. TM-score vs AlphaFold (insulin, 110 residues) = 0.0465 — below random. The model builds the chain sequentially from local geometry; it does not model long-range contacts, the hydrophobic core, disulfide bridges, or solvent.

”Masterpieces of protein architecture” series

The Nanoworld Laboratory newsletter continues the Masterpieces of protein architecture series — 193+ instalments as of April 2026 — describing structural patterns of proteins through the ring-polyhedral lens: Kushelev fractal helices, inter-turn disulfide / polysulfide bridges, ultra-long α/π helices.

What we are looking for

Researchers: independent reproduction of the validation on new datasets and independent statistical assessment of the helix signal. Biotech companies: discussion of API access and integration into codon-optimisation pipelines.