DELPHI
Screen antibody sequences for assay-specific developability liabilities.
Results
Loading starting example…
Method & evidence
Estimate PSR or SEC pass probabilities with the IPI-trained DELPHI one-hot Transformer.
Scope & limitations
- Initial profiles cover the IPI-trained paired-chain one-hot PSR and SEC models.
- Training, language-model variants, residue explanations, HIC, and viscosity are outside this initial integration.
- Local example replay proves implementation parity, not independent predictive accuracy.
Interpreting results
- Higher scores mean greater predicted probability of passing the selected assay.
- These predictions do not replace experimental developability measurements.
References
Input requirements & outputs
Input checks
- 1–32 paired antibodies
- VH and VL: 70–135 residues each
- HCDR3: 1–25 residues, IMGT 105–117 without framework anchors
- No automatic sequence truncation or HCDR3 extraction
Expected results
- Per-antibody predictions — Assay pass probability and pass/fail classification, preserving all selected antibodies.
