Protein Design Workflow Guide
Standard binder design pipeline
Overview
Target Preparation --> Backbone Generation --> Sequence Design
| | |
v v v
(pdb skill) (rfdiffusion) (proteinmpnn)
| |
v v
Structure Validation --> Filtering
| |
v v
(alphafold/chai) (protein-qc)
Phase 1: Target preparation
1.1 Obtain target structure
# Download from PDB
curl -o target.pdb "https://files.rcsb.org/download/XXXX.pdb"
1.2 Clean and prepare
# Extract target chain
# Remove waters, ligands if needed
# Trim to binding region + 10A buffer
1.3 Select hotspots
- Choose 3-6 exposed residues
- Prefer charged/aromatic (K, R, E, D, W, Y, F)
- Check surface accessibility
- Verify residue numbering
Output: target_prepared.pdb, hotspot list
Phase 2: Backbone generation
Option A: RFdiffusion (diverse exploration)
# RFdiffusion runs from the official repo, not biomodals
python run_inference.py \
inference.input_pdb=target_prepared.pdb \
contigmap.contigs=[A1-150/0 70-100] \
ppi.hotspot_res=[A45,A67,A89] \
inference.num_designs=500
Option B: BindCraft (end-to-end)
modal run modal_bindcraft.py \
--input-pdb target_prepared.pdb \
--target-hotspot-residues "45,67,89" \
--number-of-final-designs 100
Output: 100-500 backbone PDBs
Phase 3: Sequence design
For RFdiffusion backbones
for backbone in backbones/*.pdb; do
modal run modal_ligandmpnn.py \
--input-pdb "$backbone" \
--params-str "--number_of_batches 8 --temperature 0.1"
done
Output: 8 sequences per backbone (800-4000 total)
Phase 4: Structure validation
Predict complexes
# Prepare FASTA with binder + target
# binder:target format for multimer
modal run modal_alphafold.py \
--input-fasta all_sequences.fasta \
--out-dir predictions/
Output: AF2 predictions with pLDDT, ipTM, PAE
Phase 5: Filtering and selection
Apply standard thresholds
import pandas as pd
# Load metrics
designs = pd.read_csv('all_metrics.csv')
# Filter
filtered = designs[
(designs['pLDDT'] > 0.85) &
(designs['ipTM'] > 0.50) &
(designs['PAE_interface'] < 10) &
(designs['scRMSD'] < 2.0) &
(designs['esm2_pll'] > 0.0)
]
# Rank by composite score
filtered['score'] = (
0.3 * filtered['pLDDT'] +
0.3 * filtered['ipTM'] +
0.2 * (1 - filtered['PAE_interface'] / 20) +
0.2 * filtered['esm2_pll']
)
top_designs = filtered.nlargest(50, 'score')
Output: 50-200 filtered candidates
Resource planning
Compute requirements
| Stage | GPU | Time (100 designs) |
|---|
| RFdiffusion | A10G | 30 min |
| ProteinMPNN | T4 | 15 min |
| Chai / AlphaFold | A100 | 4-8 hours |
| Filtering | CPU | 15 min |
Total timeline
- Small campaign (100 designs): 8-12 hours
- Medium campaign (500 designs): 24-48 hours
- Large campaign (1000+ designs): 2-5 days
Quality checkpoints
After backbone generation
After sequence design
After validation
Final selection
Common issues
| Problem | Solution |
|---|
| Low ipTM | Check hotspots, increase designs |
| Poor diversity | Higher temperature, more backbones |
| High scRMSD | Backbone may be unusual |
| Low pLDDT | Check design quality |
Advanced workflows
Multi-tool combination
- RFdiffusion for initial backbones
- Mosaic for gradient-based refinement
- ProteinMPNN diversification
- AF2 final validation
Iterative refinement
- Run initial campaign
- Analyze failures
- Adjust hotspots/parameters
- Repeat with insights
