End-to-end computational workflow
From sequence to shortlist in 5 steps
Input a FASTA or PDB. Receive a ranked candidate list with ΔΔG scores, developability flags, and annotated structures. Single antibody: ~8 min. Batch of 50 variants: ~2 hr.
Pipeline walkthrough
Five steps from sequence to output
Input
Provide antibody sequence(s) in FASTA format — variable region (Fv) or full-length IgG. For the antigen, submit a FASTA sequence, a PDB structure file, or a homology model. Batch submissions up to 500 CDR variant sequences accepted in one job. No local software installation required.
Structure prediction
Antibody sequences are processed through an AlphaFold2 Fv prediction pipeline to generate the variable domain structure (VH + VL). The predicted Fv is modeled in canonical orientation with CDR loops (H1, H2, H3, L1, L2, L3) resolved. Where available, the antigen's existing crystal structure is used directly rather than predicted.
Interface mapping
The predicted Fv is docked against the antigen using the RosettaDock protocol. Multiple docking poses are sampled and filtered for lowest-energy configurations. Contact residues are identified by distance and energy criteria: CDR positions within 5 Å of antigen surface, hydrogen bond geometry evaluation, van der Waals contact energy at the paratope–epitope interface, and buried surface area (BSA) calculation.
Variant scoring
For each CDR variant in the batch, ΔΔG is computed relative to the wild-type using the Rosetta REF2015 energy function. The model is calibrated on SAbDab structural data and SKEMPI2 binding affinity measurements. Concurrently, all 7 developability flags are evaluated per variant: aggregation propensity (SAP), viscosity index, deamidation risk, oxidation sites, polyreactivity charge patches, FcRn-predicted half-life, and CHO expression yield.
Ranked output
Pareto-optimal candidates are selected from the affinity × developability × expression yield space. The shortlist is delivered as a ranked CSV with all per-variant scores, annotated PDB files for the top 10 candidates, and a structured JSON API response you can pipe into downstream workflows. A per-residue energy decomposition PDF is included for every top candidate.
API access
REST API — call the pipeline programmatically
POST your FASTA and antigen PDB to /v1/predict. The job runs asynchronously — poll /v1/results/{job_id} for status and results. Full reference docs available to Discovery and Campaign tier customers.
# Submit prediction job
curl -X POST https://api.genoluxo.com/v1/predict \
-H "Authorization: Bearer <YOUR_API_KEY>" \
-H "Content-Type: application/json" \
-d '{
"antibody_fasta": "EVQLVESGGGLVQPGGSLRLSCAASGFTFS...",
"antigen_pdb_url": "https://storage.example.com/target.pdb",
"cdr_variants": [
"EVQLVESGGGLVQPGGSLRLSCAASGFTFS...",
"EVQLVESGGGLVQPGGSLRLSCAASGFTFA..."
],
"run_developability": true,
"queue": "standard"
}'
# Response
{
"job_id": "jb_7f4a2c9e3d1b",
"status": "queued",
"estimated_runtime_s": 480,
"poll_url": "/v1/results/jb_7f4a2c9e3d1b"
}# Poll for results
curl https://api.genoluxo.com/v1/results/jb_7f4a2c9e3d1b \
-H "Authorization: Bearer <YOUR_API_KEY>"
# Completed response (abbreviated)
{
"status": "completed",
"candidates": [
{
"rank": 1,
"sequence_id": "VAR_003",
"ddg_kcal_mol": -2.87,
"kd_pred_nm": 0.42,
"developability_score": 0.91,
"flags_raised": 0,
"pareto_optimal": true,
"pdb_url": "/results/jb_7f4a2c9e3d1b/VAR_003.pdb"
}
]
}Turnaround times
Runtime estimates and queue SLAs
| Job type | Standard queue | Priority queue |
|---|---|---|
| Single antibody — full analysis | ~8 min | ~5 min |
| Batch of 50 CDR variants | ~2 hr | ~1 hr |
| CDR mutation scan — 500 variants | ~12 hr | ~6 hr |
| Developability-only screen (no docking) | ~2 min | ~1 min |
| Interface mapping only (no variant scoring) | ~6 min | ~4 min |
Priority queue available on Discovery tier and above. Campaign tier customers receive dedicated compute allocation with no queue position dependency.