Protein Roundtrip

PDB → Protein to Music → MIDI → Music to Protein → PDB. Test how well the dual translation preserves structural information through a full cycle.

Input Parameters

Protein Name

Sequence (DNA/RNA or protein)

PDB Structure File

If no sequence is entered, the sequence will be extracted from the PDB automatically. If no PDB is uploaded, ESMFold predicts the structure. You can use sequence-only, PDB-only, or both.

Tip for richer music: provide both a DNA/RNA sequence and a PDB file. The DNA codons add species-specific rhythmic weight (1–4 beats per residue), while the PDB structure determines the 3D neighbor-driven melody & timbre.

Species

Codon usage bias for frequency mapping

Tonality

Melody Strategy

Physical Weight

0=JazzGPT ↔ 4=Pure physics (best roundtrip)

Neighbor Cutoff (Å)

M2P ESM2 Weight

0% ESM2 · 100% physics

Beats per Grid

Positive integer. Time-signature beats grouped into each MIDI grid cell.

Random Seed (optional)

Leave empty to sample five melody environments; enter a seed for one reproducible trial.

How It Works

1. Protein → Music (P2M)
Your sequence/PDB is mapped to a jazz MIDI composition. Amino acid properties drive chord selection; 3D spatial neighbors shape timbre.

2. Music → Protein (M2P)
The generated MIDI is decoded back into a protein sequence via chord→amino-acid mapping, then folded by ESMFold into a new 3D structure.

3. Compare
The results page shows both 3D structures side by side, the audio, score notation, and a sequence alignment to quantify roundtrip fidelity.