campaign-manager

Name: campaign-manager
Author: adaptyvbio

38installs20forks

How to install campaign-manager?

adaptyvbio/protein-design-skills/campaign-manager

$npx -y skills add adaptyvbio/protein-design-skills --skill campaign-manager

Installs into the current project.

Summary

Goal-oriented binder design campaign planning and health assessment. Use this skill when: (1) Planning a complete binder design campaign, (2) Converting high-level goals into runnable pipelines, (3) Assessing campaign health and pass rates, (4) Diagnosing why designs are failing

Files · 1

View on GitHub

SKILL.md

1	# Campaign Manager
2
3	## Goal-oriented design
4
5	### From goal to pipeline
6
7	When user says: "I need 10 good binders for EGFR"
8
9	Campaign Planning:
10	```
11	Goal: 10 high-quality binders for EGFR
12	├── Achievable: Yes (standard target)
13	├── Recommended pipeline: rfdiffusion → proteinmpnn → chai → protein-qc
14	├── Estimated designs needed: 500 backbones (to get ~50 passing QC)
15	├── Estimated time: 8-12 hours total
16	├── Estimated cost: ~$60 (Modal GPU compute)
17	└── Expected yield:
18	├── After backbone (500): 500 structures
19	├── After sequence (×8): 4,000 sequences
20	├── After validation: 4,000 predictions
21	├── After QC (~10-15%): 400-600 candidates
22	└── After clustering: 10-20 diverse final designs
23	```
24
25	---
26
27	## Complete pipeline generator
28
29	### Standard miniprotein binder campaign
30
31	```bash
32	# Step 1: Fetch and prepare target (5 min)
33	curl -o target.pdb "https://files.rcsb.org/download/{PDB_ID}.pdb"
34	# Trim to binding region if needed
35
36	# Step 2: Generate backbones (2-3h, ~$15)
37	# RFdiffusion runs from the official repo, not biomodals
38	python run_inference.py \
39	inference.input_pdb=target.pdb \
40	contigmap.contigs=[A1-150/0 70-100] \
41	ppi.hotspot_res=[A45,A67,A89] \
42	inference.num_designs=500
43
44	# Checkpoint: ls output/*.pdb \| wc -l # Should be 500
45
46	# Step 3: Design sequences (1-2h, ~$10)
47	for f in output/*.pdb; do
48	modal run modal_ligandmpnn.py \
49	--input-pdb "$f" \
50	--params-str "--number_of_batches 8 --temperature 0.1"
51	done
52
53	# Checkpoint: grep -c "^>" output/seqs/*.fa # Should be ~4000
54
55	# Step 4: Quick ESM2 filter (30 min, ~$5, optional)
56	modal run modal_esm2_predict_masked.py --input-faa output/all_seqs.fa
57	# Filter sequences with PLL < 0.0
58
59	# Step 5: Structure validation (3-4h, ~$35)
60	modal run modal_alphafold.py \
61	--input-faa output/filtered_seqs.fa \
62	--out-dir predictions/
63
64	# Checkpoint: find predictions -name "*rank_001.pdb" \| wc -l
65
66	# Step 6: Filter and rank (protein-qc skill)
67	# Apply thresholds: pLDDT > 0.85, ipTM > 0.5, scRMSD < 2.0
68	# Compute composite score
69	# Cluster at 70% identity, select top from each cluster
70	```
71
72	Total estimated time: 8-12 hours
73	Total estimated cost: ~$60-70
74
75	---
76
77	## Campaign size recommendations
78
79	\| Goal \| Backbones \| Sequences/BB \| Total Seq \| Expected Passing \|
80	\|------\|-----------\|--------------\|-----------\|------------------\|
81	\| 5 binders \| 200 \| 8 \| 1,600 \| 160-240 \|
82	\| 10 binders \| 500 \| 8 \| 4,000 \| 400-600 \|
83	\| 20 binders \| 1,000 \| 8 \| 8,000 \| 800-1,200 \|
84	\| 50 binders \| 2,500 \| 8 \| 20,000 \| 2,000-3,000 \|
85
86	Rule of thumb: Generate 50x more designs than you need (10-15% pass rate × clustering).
87
88	---
89
90	## Tool selection guide
91
92	### When to use each tool
93
94	\| Scenario \| Recommended Tool \| Reason \|
95	\|----------\|------------------\|--------\|
96	\| Standard miniprotein \| RFdiffusion + ProteinMPNN \| High diversity, proven \|
97	\| Need higher success rate \| BindCraft \| Integrated design loop \|
98	\| All-atom precision needed \| BoltzGen \| Side-chain aware \|
99	\| Difficult target \| Mosaic \| Gradient, multi-model objective \|
100	\| Need fast iteration \| ESMFold2 + ESM2 \| Quick screening \|
101
102	### Target difficulty assessment
103
104	\| Indicator \| Easy Target \| Difficult Target \|
105	\|-----------\|-------------\|------------------\|
106	\| Surface type \| Concave pocket \| Flat or convex \|
107	\| Conservation \| High \| Low \|
108	\| Known binders \| Yes \| No \|
109	\| Flexibility \| Rigid \| Flexible \|
110	\| Expected pass rate \| 15-20% \| 5-10% \|
111
112	---
113
114	## Campaign health assessment
115
116	### Quick metrics check
117
118	```python
119	import pandas as pd
120
121	def assess_campaign(csv_path):
122	df = pd.read_csv(csv_path)
123
124	# Calculate pass rates
125	plddt_pass = (df['pLDDT'] > 0.85).mean()
126	iptm_pass = (df['ipTM'] > 0.50).mean()
127	scrmsd_pass = (df['scRMSD'] < 2.0).mean()
128	all_pass = ((df['pLDDT'] > 0.85) & (df['ipTM'] > 0.5) & (df['scRMSD'] < 2.0)).mean()
129
130	# Determine health
131	if all_pass > 0.15:
132	health = "EXCELLENT"
133	elif all_pass > 0.10:
134	health = "GOOD"
135	elif all_pass > 0.05:
136	health = "MARGINAL"
137	else:
138	health = "POOR"
139
140	# Identify top issue
141	issues = []
142	if plddt_pass < 0.20:
143	issues.append("Low pLDDT - backbone or sequence issue")
144	if iptm_pass < 0.20:
145	issues.append("Low ipTM - hotspot or interface issue")
146	if scrmsd_pass < 0.50:
147	issues.append("High scRMSD - sequence doesn't specify backbone")
148
149	return {
150	"health": health,

People who liked this also viewed

6 picks

Type

campaign-manager

38installs20forks

View on GitHub

How to install campaign-manager?

adaptyvbio/protein-design-skills/campaign-manager

$npx -y skills add adaptyvbio/protein-design-skills --skill campaign-manager

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Campaign Manager
2
3	## Goal-oriented design
4
5	### From goal to pipeline
6
7	When user says: "I need 10 good binders for EGFR"
8
9	Campaign Planning:
10	```
11	Goal: 10 high-quality binders for EGFR
12	├── Achievable: Yes (standard target)
13	├── Recommended pipeline: rfdiffusion → proteinmpnn → chai → protein-qc
14	├── Estimated designs needed: 500 backbones (to get ~50 passing QC)
15	├── Estimated time: 8-12 hours total
16	├── Estimated cost: ~$60 (Modal GPU compute)
17	└── Expected yield:
18	├── After backbone (500): 500 structures
19	├── After sequence (×8): 4,000 sequences
20	├── After validation: 4,000 predictions
21	├── After QC (~10-15%): 400-600 candidates
22	└── After clustering: 10-20 diverse final designs
23	```
24
25	---
26
27	## Complete pipeline generator
28
29	### Standard miniprotein binder campaign
30
31	```bash
32	# Step 1: Fetch and prepare target (5 min)
33	curl -o target.pdb "https://files.rcsb.org/download/{PDB_ID}.pdb"
34	# Trim to binding region if needed
35
36	# Step 2: Generate backbones (2-3h, ~$15)
37	# RFdiffusion runs from the official repo, not biomodals
38	python run_inference.py \
39	inference.input_pdb=target.pdb \
40	contigmap.contigs=[A1-150/0 70-100] \
41	ppi.hotspot_res=[A45,A67,A89] \
42	inference.num_designs=500
43
44	# Checkpoint: ls output/*.pdb \| wc -l # Should be 500
45
46	# Step 3: Design sequences (1-2h, ~$10)
47	for f in output/*.pdb; do
48	modal run modal_ligandmpnn.py \
49	--input-pdb "$f" \
50	--params-str "--number_of_batches 8 --temperature 0.1"
51	done
52
53	# Checkpoint: grep -c "^>" output/seqs/*.fa # Should be ~4000
54
55	# Step 4: Quick ESM2 filter (30 min, ~$5, optional)
56	modal run modal_esm2_predict_masked.py --input-faa output/all_seqs.fa
57	# Filter sequences with PLL < 0.0
58
59	# Step 5: Structure validation (3-4h, ~$35)
60	modal run modal_alphafold.py \
61	--input-faa output/filtered_seqs.fa \
62	--out-dir predictions/
63
64	# Checkpoint: find predictions -name "*rank_001.pdb" \| wc -l
65
66	# Step 6: Filter and rank (protein-qc skill)
67	# Apply thresholds: pLDDT > 0.85, ipTM > 0.5, scRMSD < 2.0
68	# Compute composite score
69	# Cluster at 70% identity, select top from each cluster
70	```
71
72	Total estimated time: 8-12 hours
73	Total estimated cost: ~$60-70
74
75	---
76
77	## Campaign size recommendations
78
79	\| Goal \| Backbones \| Sequences/BB \| Total Seq \| Expected Passing \|
80	\|------\|-----------\|--------------\|-----------\|------------------\|
81	\| 5 binders \| 200 \| 8 \| 1,600 \| 160-240 \|
82	\| 10 binders \| 500 \| 8 \| 4,000 \| 400-600 \|
83	\| 20 binders \| 1,000 \| 8 \| 8,000 \| 800-1,200 \|
84	\| 50 binders \| 2,500 \| 8 \| 20,000 \| 2,000-3,000 \|
85
86	Rule of thumb: Generate 50x more designs than you need (10-15% pass rate × clustering).
87
88	---
89
90	## Tool selection guide
91
92	### When to use each tool
93
94	\| Scenario \| Recommended Tool \| Reason \|
95	\|----------\|------------------\|--------\|
96	\| Standard miniprotein \| RFdiffusion + ProteinMPNN \| High diversity, proven \|
97	\| Need higher success rate \| BindCraft \| Integrated design loop \|
98	\| All-atom precision needed \| BoltzGen \| Side-chain aware \|
99	\| Difficult target \| Mosaic \| Gradient, multi-model objective \|
100	\| Need fast iteration \| ESMFold2 + ESM2 \| Quick screening \|
101
102	### Target difficulty assessment
103
104	\| Indicator \| Easy Target \| Difficult Target \|
105	\|-----------\|-------------\|------------------\|
106	\| Surface type \| Concave pocket \| Flat or convex \|
107	\| Conservation \| High \| Low \|
108	\| Known binders \| Yes \| No \|
109	\| Flexibility \| Rigid \| Flexible \|
110	\| Expected pass rate \| 15-20% \| 5-10% \|
111
112	---
113
114	## Campaign health assessment
115
116	### Quick metrics check
117
118	```python
119	import pandas as pd
120
121	def assess_campaign(csv_path):
122	df = pd.read_csv(csv_path)
123
124	# Calculate pass rates
125	plddt_pass = (df['pLDDT'] > 0.85).mean()
126	iptm_pass = (df['ipTM'] > 0.50).mean()
127	scrmsd_pass = (df['scRMSD'] < 2.0).mean()
128	all_pass = ((df['pLDDT'] > 0.85) & (df['ipTM'] > 0.5) & (df['scRMSD'] < 2.0)).mean()
129
130	# Determine health
131	if all_pass > 0.15:
132	health = "EXCELLENT"
133	elif all_pass > 0.10:
134	health = "GOOD"
135	elif all_pass > 0.05:
136	health = "MARGINAL"
137	else:
138	health = "POOR"
139
140	# Identify top issue
141	issues = []
142	if plddt_pass < 0.20:
143	issues.append("Low pLDDT - backbone or sequence issue")
144	if iptm_pass < 0.20:
145	issues.append("Low ipTM - hotspot or interface issue")
146	if scrmsd_pass < 0.50:
147	issues.append("High scRMSD - sequence doesn't specify backbone")
148
149	return {
150	"health": health,

People who liked this also viewed

6 picks

Type