Binder Design Campaign Manager
Plain-language role: Use this skill to plan design batches, estimate yields, and diagnose why a campaign is underperforming.
Goal-oriented design
From goal to pipeline
When user says: "I need 10 good binders for EGFR"
Campaign Planning:
Goal: 10 high-quality binders for EGFR
├── Achievable: Yes (standard target)
├── Recommended pipeline: rfdiffusion → proteinmpnn → colabfold → protein-design-qc
├── Estimated designs needed: 500 backbones (to get ~50 passing QC)
├── Estimated time: 8-12 hours total
├── Estimated cost: ~$60 (Modal GPU compute)
└── Expected yield:
├── After backbone (500): 500 structures
├── After sequence (×8): 4,000 sequences
├── After validation: 4,000 predictions
├── After QC (~10-15%): 400-600 candidates
└── After clustering: 10-20 diverse final designs
Complete pipeline generator
Standard miniprotein binder campaign
# Step 1: Fetch and prepare target (5 min)
curl -o target.pdb "https://files.rcsb.org/download/{PDB_ID}.pdb"
# Trim to binding region if needed
# Step 2: Generate backbones (2-3h, ~$15)
modal run modal_rfdiffusion.py \
--pdb target.pdb \
--contigs "A1-150/0 70-100" \
--hotspot "A45,A67,A89" \
--num-designs 500
# Checkpoint: ls output/*.pdb | wc -l # Should be 500
# Step 3: Design sequences (1-2h, ~$10)
for f in output/*.pdb; do
modal run modal_proteinmpnn.py \
--pdb-path "$f" \
--num-seq-per-target 8 \
--sampling-temp 0.1
done
# Checkpoint: grep -c "^>" output/seqs/*.fa # Should be ~4000
# Step 4: Quick ESM2 filter (30 min, ~$5, optional)
modal run modal_esm.py --fasta output/all_seqs.fa --mode pll
# Filter sequences with PLL < 0.0
# Step 5: Structure validation (3-4h, ~$35)
modal run modal_colabfold.py \
--input-faa output/filtered_seqs.fa \
--out-dir predictions/
# Checkpoint: find predictions -name "*rank_001.pdb" | wc -l
# Step 6: Filter and rank (protein-design-qc skill)
# Apply thresholds: pLDDT > 0.85, ipTM > 0.5, scRMSD < 2.0
# Compute composite score
# Cluster at 70% identity, select top from each cluster
Total estimated time: 8-12 hours
Total estimated cost: ~$60-70
Campaign size recommendations
| Goal |
Backbones |
Sequences/BB |
Total Seq |
Expected Passing |
| 5 binders |
200 |
8 |
1,600 |
160-240 |
| 10 binders |
500 |
8 |
4,000 |
400-600 |
| 20 binders |
1,000 |
8 |
8,000 |
800-1,200 |
| 50 binders |
2,500 |
8 |
20,000 |
2,000-3,000 |
Rule of thumb: Generate 50x more designs than you need (10-15% pass rate × clustering).
Tool selection guide
When to use each tool
| Scenario |
Recommended Tool |
Reason |
| Standard miniprotein |
RFdiffusion + ProteinMPNN |
High diversity, proven |
| Need higher success rate |
BindCraft |
Integrated design loop |
| All-atom precision needed |
BoltzGen |
Side-chain aware |
| Difficult target |
ColabDesign |
AF2 gradient optimization |
| Need fast iteration |
ESMFold + ESM2 |
Quick screening |
Target difficulty assessment
| Indicator |
Easy Target |
Difficult Target |
| Surface type |
Concave pocket |
Flat or convex |
| Conservation |
High |
Low |
| Known binders |
Yes |
No |
| Flexibility |
Rigid |
Flexible |
| Expected pass rate |
15-20% |
5-10% |
Campaign health assessment
Quick metrics check
import pandas as pd
def assess_campaign(csv_path):
df = pd.read_csv(csv_path)
# Calculate pass rates
plddt_pass = (df['pLDDT'] > 0.85).mean()
iptm_pass = (df['ipTM'] > 0.50).mean()
scrmsd_pass = (df['scRMSD'] < 2.0).mean()
all_pass = ((df['pLDDT'] > 0.85) & (df['ipTM'] > 0.5) & (df['scRMSD'] < 2.0)).mean()
# Determine health
if all_pass > 0.15:
health = "EXCELLENT"
elif all_pass > 0.10:
health = "GOOD"
elif all_pass > 0.05:
health = "MARGINAL"
else:
health = "POOR"
# Identify top issue
issues = []
if plddt_pass < 0.20:
issues.append("Low pLDDT - backbone or sequence issue")
if iptm_pass < 0.20:
issues.append("Low ipTM - hotspot or interface issue")
if scrmsd_pass < 0.50:
issues.append("High scRMSD - sequence doesn't specify backbone")
return {
"health": health,
"overall_pass_rate": all_pass,
"plddt_pass_rate": plddt_pass,
"iptm_pass_rate": iptm_pass,
"scrmsd_pass_rate": scrmsd_pass,
"top_issues": issues
}
Interpreting results
| Health |
Pass Rate |
Action |
| EXCELLENT |
> 15% |
Proceed to selection |
| GOOD |
10-15% |
Proceed, normal yield |
| MARGINAL |
5-10% |
Review failure tree |
| POOR |
< 5% |
Diagnose and restart |
Cost estimation
Per-tool costs (Modal)
| Tool |
GPU |
$/hour |
Typical Job |
Cost |
| RFdiffusion |
A10G |
~$1.20 |
500 designs/2h |
~$2.50 |
| ProteinMPNN |
T4 |
~$0.60 |
4000 seq/1.5h |
~$1.00 |
| ESM2 (PLL) |
A10G |
~$1.20 |
4000 seq/30min |
~$0.60 |
| ColabFold |
A100 |
~$4.50 |
4000 preds/4h |
~$18.00 |
| Chai |
A100 |
~$4.50 |
500 preds/1h |
~$4.50 |
Campaign cost estimates
| Campaign Size |
Total Cost |
Notes |
| Small (100 bb) |
~$15 |
Quick exploration |
| Standard (500 bb) |
~$60 |
Most campaigns |
| Large (1000 bb) |
~$120 |
Comprehensive |
| XL (5000 bb) |
~$600 |
Very thorough |
Pipeline variants
High-throughput (maximize diversity)
# More backbones, fewer sequences each
modal run modal_rfdiffusion.py --num-designs 2000
modal run modal_proteinmpnn.py --num-seq-per-target 4 --sampling-temp 0.2
High-quality (maximize per-design quality)
# Fewer backbones, more sequences each, lower temperature
modal run modal_rfdiffusion.py --num-designs 200
modal run modal_proteinmpnn.py --num-seq-per-target 32 --sampling-temp 0.1
Quick exploration (fast iteration)
# Small batch, ESMFold instead of ColabFold
modal run modal_rfdiffusion.py --num-designs 50
modal run modal_proteinmpnn.py --num-seq-per-target 8
modal run modal_esmfold.py --fasta all_seqs.fa # Faster than ColabFold
See also
- Tool-specific parameters:
rfdiffusion, proteinmpnn, colabfold, chai1-structure-prediction, boltz-structure-prediction
- QC thresholds and filtering:
protein-design-qc
- Tool selection guidance:
binder-design-tool-selection
Inputs
- A campaign goal expressed in experimental terms such as binder count, timeline, or target class.
- Upstream context including target structure quality, hotspot knowledge, and compute budget.
- Observed campaign metrics if the design effort is already in progress.
Outputs
- A concrete campaign plan with tool selection, design counts, and expected funnel sizes.
- Health checks on failure modes such as poor pLDDT, weak interfaces, or low diversity.
- Prioritized next actions that map into the relevant generation, validation, or QC skills.
Next Step
Execute the planned campaign through binder-design-tool-selection and end-to-end-protein-design-workflow, then revisit this skill when pass rates drift.
1---
2name: binder-design-campaign-manager
3description: >
4 Binder design campaign planning, monitoring, and troubleshooting.
5 Use this skill when: (1) Planning a complete binder design campaign,
6 (2) Converting high-level goals into runnable pipelines,
7 (3) Assessing campaign health and pass rates,
8 (4) Diagnosing why designs are failing QC,
9 (5) Estimating time, cost, and expected yields,
10 (6) Selecting between design tools for a specific target.
11
12 This skill orchestrates the other protein design tools.
13 For individual tool parameters, use the specific tool skills.
14license: MIT
15category: orchestration
16tags: [planning, campaign, coordination]
17metadata:
18 source: https://skillmd.com/skills/biotender-max/binder-design-campaign-manager
19---
20
21# Binder Design Campaign Manager
22
23**Plain-language role**: Use this skill to plan design batches, estimate yields, and diagnose why a campaign is underperforming.
24
25## Goal-oriented design
26
27### From goal to pipeline
28
29When user says: "I need 10 good binders for EGFR"
30
31**Campaign Planning:**
32```
33Goal: 10 high-quality binders for EGFR
34├── Achievable: Yes (standard target)
35├── Recommended pipeline: rfdiffusion → proteinmpnn → colabfold → protein-design-qc
36├── Estimated designs needed: 500 backbones (to get ~50 passing QC)
37├── Estimated time: 8-12 hours total
38├── Estimated cost: ~$60 (Modal GPU compute)
39└── Expected yield:
40 ├── After backbone (500): 500 structures
41 ├── After sequence (×8): 4,000 sequences
42 ├── After validation: 4,000 predictions
43 ├── After QC (~10-15%): 400-600 candidates
44 └── After clustering: 10-20 diverse final designs
45```
46
47---
48
49## Complete pipeline generator
50
51### Standard miniprotein binder campaign
52
53```bash
54# Step 1: Fetch and prepare target (5 min)
55curl -o target.pdb "https://files.rcsb.org/download/{PDB_ID}.pdb"
56# Trim to binding region if needed
57
58# Step 2: Generate backbones (2-3h, ~$15)
59modal run modal_rfdiffusion.py \
60 --pdb target.pdb \
61 --contigs "A1-150/0 70-100" \
62 --hotspot "A45,A67,A89" \
63 --num-designs 500
64
65# Checkpoint: ls output/*.pdb | wc -l # Should be 500
66
67# Step 3: Design sequences (1-2h, ~$10)
68for f in output/*.pdb; do
69 modal run modal_proteinmpnn.py \
70 --pdb-path "$f" \
71 --num-seq-per-target 8 \
72 --sampling-temp 0.1
73done
74
75# Checkpoint: grep -c "^>" output/seqs/*.fa # Should be ~4000
76
77# Step 4: Quick ESM2 filter (30 min, ~$5, optional)
78modal run modal_esm.py --fasta output/all_seqs.fa --mode pll
79# Filter sequences with PLL < 0.0
80
81# Step 5: Structure validation (3-4h, ~$35)
82modal run modal_colabfold.py \
83 --input-faa output/filtered_seqs.fa \
84 --out-dir predictions/
85
86# Checkpoint: find predictions -name "*rank_001.pdb" | wc -l
87
88# Step 6: Filter and rank (protein-design-qc skill)
89# Apply thresholds: pLDDT > 0.85, ipTM > 0.5, scRMSD < 2.0
90# Compute composite score
91# Cluster at 70% identity, select top from each cluster
92```
93
94**Total estimated time**: 8-12 hours
95**Total estimated cost**: ~$60-70
96
97---
98
99## Campaign size recommendations
100
101| Goal | Backbones | Sequences/BB | Total Seq | Expected Passing |
102|------|-----------|--------------|-----------|------------------|
103| 5 binders | 200 | 8 | 1,600 | 160-240 |
104| 10 binders | 500 | 8 | 4,000 | 400-600 |
105| 20 binders | 1,000 | 8 | 8,000 | 800-1,200 |
106| 50 binders | 2,500 | 8 | 20,000 | 2,000-3,000 |
107
108**Rule of thumb**: Generate 50x more designs than you need (10-15% pass rate × clustering).
109
110---
111
112## Tool selection guide
113
114### When to use each tool
115
116| Scenario | Recommended Tool | Reason |
117|----------|------------------|--------|
118| Standard miniprotein | RFdiffusion + ProteinMPNN | High diversity, proven |
119| Need higher success rate | BindCraft | Integrated design loop |
120| All-atom precision needed | BoltzGen | Side-chain aware |
121| Difficult target | ColabDesign | AF2 gradient optimization |
122| Need fast iteration | ESMFold + ESM2 | Quick screening |
123
124### Target difficulty assessment
125
126| Indicator | Easy Target | Difficult Target |
127|-----------|-------------|------------------|
128| Surface type | Concave pocket | Flat or convex |
129| Conservation | High | Low |
130| Known binders | Yes | No |
131| Flexibility | Rigid | Flexible |
132| Expected pass rate | 15-20% | 5-10% |
133
134---
135
136## Campaign health assessment
137
138### Quick metrics check
139
140```python
141import pandas as pd
142
143def assess_campaign(csv_path):
144 df = pd.read_csv(csv_path)
145
146 # Calculate pass rates
147 plddt_pass = (df['pLDDT'] > 0.85).mean()
148 iptm_pass = (df['ipTM'] > 0.50).mean()
149 scrmsd_pass = (df['scRMSD'] < 2.0).mean()
150 all_pass = ((df['pLDDT'] > 0.85) & (df['ipTM'] > 0.5) & (df['scRMSD'] < 2.0)).mean()
151
152 # Determine health
153 if all_pass > 0.15:
154 health = "EXCELLENT"
155 elif all_pass > 0.10:
156 health = "GOOD"
157 elif all_pass > 0.05:
158 health = "MARGINAL"
159 else:
160 health = "POOR"
161
162 # Identify top issue
163 issues = []
164 if plddt_pass < 0.20:
165 issues.append("Low pLDDT - backbone or sequence issue")
166 if iptm_pass < 0.20:
167 issues.append("Low ipTM - hotspot or interface issue")
168 if scrmsd_pass < 0.50:
169 issues.append("High scRMSD - sequence doesn't specify backbone")
170
171 return {
172 "health": health,
173 "overall_pass_rate": all_pass,
174 "plddt_pass_rate": plddt_pass,
175 "iptm_pass_rate": iptm_pass,
176 "scrmsd_pass_rate": scrmsd_pass,
177 "top_issues": issues
178 }
179```
180
181### Interpreting results
182
183| Health | Pass Rate | Action |
184|--------|-----------|--------|
185| EXCELLENT | > 15% | Proceed to selection |
186| GOOD | 10-15% | Proceed, normal yield |
187| MARGINAL | 5-10% | Review failure tree |
188| POOR | < 5% | Diagnose and restart |
189
190---
191
192## Cost estimation
193
194### Per-tool costs (Modal)
195
196| Tool | GPU | $/hour | Typical Job | Cost |
197|------|-----|--------|-------------|------|
198| RFdiffusion | A10G | ~$1.20 | 500 designs/2h | ~$2.50 |
199| ProteinMPNN | T4 | ~$0.60 | 4000 seq/1.5h | ~$1.00 |
200| ESM2 (PLL) | A10G | ~$1.20 | 4000 seq/30min | ~$0.60 |
201| ColabFold | A100 | ~$4.50 | 4000 preds/4h | ~$18.00 |
202| Chai | A100 | ~$4.50 | 500 preds/1h | ~$4.50 |
203
204### Campaign cost estimates
205
206| Campaign Size | Total Cost | Notes |
207|---------------|------------|-------|
208| Small (100 bb) | ~$15 | Quick exploration |
209| Standard (500 bb) | ~$60 | Most campaigns |
210| Large (1000 bb) | ~$120 | Comprehensive |
211| XL (5000 bb) | ~$600 | Very thorough |
212
213---
214
215## Pipeline variants
216
217### High-throughput (maximize diversity)
218
219```bash
220# More backbones, fewer sequences each
221modal run modal_rfdiffusion.py --num-designs 2000
222modal run modal_proteinmpnn.py --num-seq-per-target 4 --sampling-temp 0.2
223```
224
225### High-quality (maximize per-design quality)
226
227```bash
228# Fewer backbones, more sequences each, lower temperature
229modal run modal_rfdiffusion.py --num-designs 200
230modal run modal_proteinmpnn.py --num-seq-per-target 32 --sampling-temp 0.1
231```
232
233### Quick exploration (fast iteration)
234
235```bash
236# Small batch, ESMFold instead of ColabFold
237modal run modal_rfdiffusion.py --num-designs 50
238modal run modal_proteinmpnn.py --num-seq-per-target 8
239modal run modal_esmfold.py --fasta all_seqs.fa # Faster than ColabFold
240```
241
242---
243
244## See also
245
246- Tool-specific parameters: `rfdiffusion`, `proteinmpnn`, `colabfold`, `chai1-structure-prediction`, `boltz-structure-prediction`
247- QC thresholds and filtering: `protein-design-qc`
248- Tool selection guidance: `binder-design-tool-selection`
249
250## Inputs
251
252- A campaign goal expressed in experimental terms such as binder count, timeline, or target class.
253- Upstream context including target structure quality, hotspot knowledge, and compute budget.
254- Observed campaign metrics if the design effort is already in progress.
255
256## Outputs
257
258- A concrete campaign plan with tool selection, design counts, and expected funnel sizes.
259- Health checks on failure modes such as poor pLDDT, weak interfaces, or low diversity.
260- Prioritized next actions that map into the relevant generation, validation, or QC skills.
261
262## Next Step
263
264Execute the planned campaign through `binder-design-tool-selection` and `end-to-end-protein-design-workflow`, then revisit this skill when pass rates drift.