joelinator/dflow-novo-model

🧬 DFlowNovo: Discrete Flow Matching for De Novo Peptide Sequencing

DFlowNovo is a state-of-the-art de novo peptide sequencing framework powered by Discrete Flow Matching (DFM) with full support for Post-Translational Modifications (PTMs) and UNIMOD standard formatting.

---

🚀 Key Features

• Discrete Flow Matching: Continuous-time generative modeling on discrete token sequences for accurate and coherent peptide reconstruction.
• Top-k Length Bayesian Beam Decoding: Dynamic length exploration over the top predicted candidate lengths ranked by joint posterior likelihood.
• Post-Translational Modifications (PTMs):
  • Methionine Oxidation: M(ox) / M[UNIMOD:35] (+15.9949 Da)
  • Cysteine Carboxyamidomethylation: C(cam) / C[UNIMOD:4] (+57.0215 Da)
  • Asparagine / Glutamine Deamidation: N(deam) / Q(deam) / [UNIMOD:7] (+0.9840 Da)
  • Serine / Threonine / Tyrosine Phosphorylation: S(ph) / T(ph) / Y(ph) / [UNIMOD:21] (+79.9663 Da)
  • N-terminal Acetylation: (+42.01) / [UNIMOD:1] (+42.0106 Da)
  • N-terminal Carbamylation: (+43.01) / [UNIMOD:5] (+43.0058 Da)
  • N-terminal Ammonia Loss: (-17.03) / [UNIMOD:385] (-17.0265 Da)
• Input Formats: Mascot Generic Format (.mgf), Compressed MGF (.mgz / .mgf.gz).
• Interactive Outputs: Real-time tabular output with confidence scores, precursor and calculated peptide masses, mass errors in ppm, and export to CSV/TSV.

---

📖 How to Use

1. Upload an .mgz or .mgf file containing MS/MS spectra (or click the example file).
2. Adjust inference parameters (number of integration steps, length candidates, guidance scale) if desired.
3. Click "Run De Novo Sequencing".
4. View and download the predicted peptide sequences and spectrum metrics in the interactive results table.