bindcraft

Name: bindcraft
Author: adaptyvbio

30installs20forks

How to install bindcraft?

adaptyvbio/protein-design-skills/bindcraft

$npx -y skills add adaptyvbio/protein-design-skills --skill bindcraft

Installs into the current project.

Summary

End-to-end binder design using BindCraft hallucination. Use this skill when: (1) Designing protein binders with built-in AF2 validation, (2) Running production-quality binder campaigns, (3) Using different design protocols (fast, default, slow), (4) Need joint backbone and sequen

Files · 1

View on GitHub

SKILL.md

1	# BindCraft Binder Design
2
3	## Prerequisites
4
5	\| Requirement \| Minimum \| Recommended \|
6	\|-------------\|---------\|-------------\|
7	\| Python \| 3.9+ \| 3.10 \|
8	\| CUDA \| 11.7+ \| 12.0+ \|
9	\| GPU VRAM \| 32GB \| 48GB (L40S) \|
10	\| RAM \| 32GB \| 64GB \|
11
12	## How to run
13
14	> First time? See [Getting started](../../docs/getting-started.md) to set up Modal and biomodals.
15
16	### Option 1: Modal (recommended)
17	```bash
18	cd biomodals
19	modal run modal_bindcraft.py \
20	--input-pdb target.pdb \
21	--target-chains A \
22	--target-hotspot-residues "45,67,89" \
23	--lengths "70,100" \
24	--number-of-final-designs 50
25	```
26
27	GPU: L40S (48GB) \| Timeout: 300 min default
28
29	### Option 2: Local installation
30	```bash
31	git clone https://github.com/martinpacesa/BindCraft.git
32	cd BindCraft
33
34	# BindCraft is configured with JSON files, not flags
35	python -u ./bindcraft.py \
36	--settings ./settings_target/mytarget.json \
37	--filters ./settings_filters/default_filters.json \
38	--advanced ./settings_advanced/default_4stage_multimer.json
39	```
40
41	The target PDB, chains, hotspots, and binder length range are set inside the
42	`--settings` JSON. See the BindCraft repo for the settings schema.
43
44	## Key parameters (Modal wrapper)
45
46	\| Parameter \| Default \| Description \|
47	\|-----------\|---------\|-------------\|
48	\| `--input-pdb` \| required \| Target structure \|
49	\| `--target-chains` \| `A` \| Target chain(s) \|
50	\| `--target-hotspot-residues` \| "" \| Target hotspots (e.g. "45,67,89") \|
51	\| `--lengths` \| `50,130` \| Binder length range \|
52	\| `--number-of-final-designs` \| 1 \| Passing designs to return \|
53	\| `--max-trajectories` \| none \| Cap on trajectories \|
54
55	## Output format
56
57	```
58	output/
59	├── design_0/
60	│ ├── binder.pdb # Final design
61	│ ├── complex.pdb # Binder + target
62	│ ├── metrics.json # QC scores
63	│ └── trajectory/ # Optimization trajectory
64	├── design_1/
65	│ └── ...
66	└── summary.csv # All metrics
67	```
68
69	### Metrics Output
70	```json
71	{
72	"plddt": 0.89,
73	"ptm": 0.78,
74	"iptm": 0.62,
75	"pae": 8.5,
76	"rmsd": 1.2,
77	"sequence": "MKTAYIAK..."
78	}
79	```
80
81	## Sample output
82
83	### Successful run
84	```
85	$ modal run modal_bindcraft.py --input-pdb target.pdb --target-chains A --target-hotspot-residues "45,67,89" --number-of-final-designs 50
86	[INFO] Loading BindCraft model...
87	[INFO] Target: target.pdb (chain A)
88	[INFO] Hotspots: 45, 67, 89
89	[INFO] Generating designs...
90
91	Design 1/50:
92	Length: 78 AA
93	pLDDT: 0.89, ipTM: 0.62
94	Saved: output/design_0/
95
96	Design 50/50:
97	Length: 85 AA
98	pLDDT: 0.86, ipTM: 0.58
99	Saved: output/design_49/
100
101	[INFO] Campaign complete. Summary: output/summary.csv
102	Pass rate: 32/50 (64%) with ipTM > 0.5
103	```
104
105	What good output looks like:
106	- pLDDT: > 0.85 for most designs
107	- ipTM: > 0.5 for passing designs
108	- Pass rate: 30-70% depending on target
109	- Diverse sequences across designs
110
111	## Decision tree
112
113	```
114	Should I use BindCraft?
115	│
116	├─ What type of design?
117	│ ├─ Production-quality binders → BindCraft ✓
118	│ ├─ High diversity exploration → RFdiffusion
119	│ └─ All-atom precision → BoltzGen
120	│
121	├─ What matters most?
122	│ ├─ Experimental success rate → BindCraft ✓
123	│ ├─ Speed / diversity → RFdiffusion + ProteinMPNN
124	│ ├─ AF2 gradient optimization → ColabDesign
125	│ └─ All-atom control → BoltzGen
126	│
127	└─ Compute resources?
128	├─ Have L40S/A100 → BindCraft ✓
129	└─ Only A10G → RFdiffusion + ProteinMPNN
130	```
131
132	## Typical performance
133
134	\| Campaign Size \| Time (L40S) \| Cost (Modal) \| Notes \|
135	\|---------------\|-------------\|--------------\|-------\|
136	\| 50 designs \| 2-4h \| ~$15 \| Quick campaign \|
137	\| 100 designs \| 4-8h \| ~$30 \| Standard \|
138	\| 200 designs \| 8-16h \| ~$60 \| Large campaign \|
139
140	Adaptyv's own tests of these models showed BindCraft costing about $2.90 per accepted
141	design, averaged across 7 targets.
142
143	Experimental success rate (BindCraft paper): 10 to 100%, averaging 46.3% across 12
144	targets; strongly target-dependent.
145
146	---
147
148	## Verify
149
150	```bash
151	find output -name "binder.pdb" \| wc -l # Should match num_designs
152	```
153
154	---
155
156	## Troubleshooting
157
158	Low ipTM scores: Check hotspot selection, increase designs
159	Slow convergence: Use fast protocol for screening
160	OOM errors: Reduce num_models, use L40S GPU
161	Poor diversity: Lower sampling_temp, run multiple seeds
162
163	### Error interpretation
164
165	\| Error \| Cause \| Fix \|
166	\|-------\|-------\|-----\|
167	\| `RuntimeError: CUDA out

People who liked this also viewed

6 picks

Type

bindcraft

30installs20forks

View on GitHub

How to install bindcraft?

adaptyvbio/protein-design-skills/bindcraft

$npx -y skills add adaptyvbio/protein-design-skills --skill bindcraft

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# BindCraft Binder Design
2
3	## Prerequisites
4
5	\| Requirement \| Minimum \| Recommended \|
6	\|-------------\|---------\|-------------\|
7	\| Python \| 3.9+ \| 3.10 \|
8	\| CUDA \| 11.7+ \| 12.0+ \|
9	\| GPU VRAM \| 32GB \| 48GB (L40S) \|
10	\| RAM \| 32GB \| 64GB \|
11
12	## How to run
13
14	> First time? See [Getting started](../../docs/getting-started.md) to set up Modal and biomodals.
15
16	### Option 1: Modal (recommended)
17	```bash
18	cd biomodals
19	modal run modal_bindcraft.py \
20	--input-pdb target.pdb \
21	--target-chains A \
22	--target-hotspot-residues "45,67,89" \
23	--lengths "70,100" \
24	--number-of-final-designs 50
25	```
26
27	GPU: L40S (48GB) \| Timeout: 300 min default
28
29	### Option 2: Local installation
30	```bash
31	git clone https://github.com/martinpacesa/BindCraft.git
32	cd BindCraft
33
34	# BindCraft is configured with JSON files, not flags
35	python -u ./bindcraft.py \
36	--settings ./settings_target/mytarget.json \
37	--filters ./settings_filters/default_filters.json \
38	--advanced ./settings_advanced/default_4stage_multimer.json
39	```
40
41	The target PDB, chains, hotspots, and binder length range are set inside the
42	`--settings` JSON. See the BindCraft repo for the settings schema.
43
44	## Key parameters (Modal wrapper)
45
46	\| Parameter \| Default \| Description \|
47	\|-----------\|---------\|-------------\|
48	\| `--input-pdb` \| required \| Target structure \|
49	\| `--target-chains` \| `A` \| Target chain(s) \|
50	\| `--target-hotspot-residues` \| "" \| Target hotspots (e.g. "45,67,89") \|
51	\| `--lengths` \| `50,130` \| Binder length range \|
52	\| `--number-of-final-designs` \| 1 \| Passing designs to return \|
53	\| `--max-trajectories` \| none \| Cap on trajectories \|
54
55	## Output format
56
57	```
58	output/
59	├── design_0/
60	│ ├── binder.pdb # Final design
61	│ ├── complex.pdb # Binder + target
62	│ ├── metrics.json # QC scores
63	│ └── trajectory/ # Optimization trajectory
64	├── design_1/
65	│ └── ...
66	└── summary.csv # All metrics
67	```
68
69	### Metrics Output
70	```json
71	{
72	"plddt": 0.89,
73	"ptm": 0.78,
74	"iptm": 0.62,
75	"pae": 8.5,
76	"rmsd": 1.2,
77	"sequence": "MKTAYIAK..."
78	}
79	```
80
81	## Sample output
82
83	### Successful run
84	```
85	$ modal run modal_bindcraft.py --input-pdb target.pdb --target-chains A --target-hotspot-residues "45,67,89" --number-of-final-designs 50
86	[INFO] Loading BindCraft model...
87	[INFO] Target: target.pdb (chain A)
88	[INFO] Hotspots: 45, 67, 89
89	[INFO] Generating designs...
90
91	Design 1/50:
92	Length: 78 AA
93	pLDDT: 0.89, ipTM: 0.62
94	Saved: output/design_0/
95
96	Design 50/50:
97	Length: 85 AA
98	pLDDT: 0.86, ipTM: 0.58
99	Saved: output/design_49/
100
101	[INFO] Campaign complete. Summary: output/summary.csv
102	Pass rate: 32/50 (64%) with ipTM > 0.5
103	```
104
105	What good output looks like:
106	- pLDDT: > 0.85 for most designs
107	- ipTM: > 0.5 for passing designs
108	- Pass rate: 30-70% depending on target
109	- Diverse sequences across designs
110
111	## Decision tree
112
113	```
114	Should I use BindCraft?
115	│
116	├─ What type of design?
117	│ ├─ Production-quality binders → BindCraft ✓
118	│ ├─ High diversity exploration → RFdiffusion
119	│ └─ All-atom precision → BoltzGen
120	│
121	├─ What matters most?
122	│ ├─ Experimental success rate → BindCraft ✓
123	│ ├─ Speed / diversity → RFdiffusion + ProteinMPNN
124	│ ├─ AF2 gradient optimization → ColabDesign
125	│ └─ All-atom control → BoltzGen
126	│
127	└─ Compute resources?
128	├─ Have L40S/A100 → BindCraft ✓
129	└─ Only A10G → RFdiffusion + ProteinMPNN
130	```
131
132	## Typical performance
133
134	\| Campaign Size \| Time (L40S) \| Cost (Modal) \| Notes \|
135	\|---------------\|-------------\|--------------\|-------\|
136	\| 50 designs \| 2-4h \| ~$15 \| Quick campaign \|
137	\| 100 designs \| 4-8h \| ~$30 \| Standard \|
138	\| 200 designs \| 8-16h \| ~$60 \| Large campaign \|
139
140	Adaptyv's own tests of these models showed BindCraft costing about $2.90 per accepted
141	design, averaged across 7 targets.
142
143	Experimental success rate (BindCraft paper): 10 to 100%, averaging 46.3% across 12
144	targets; strongly target-dependent.
145
146	---
147
148	## Verify
149
150	```bash
151	find output -name "binder.pdb" \| wc -l # Should match num_designs
152	```
153
154	---
155
156	## Troubleshooting
157
158	Low ipTM scores: Check hotspot selection, increase designs
159	Slow convergence: Use fast protocol for screening
160	OOM errors: Reduce num_models, use L40S GPU
161	Poor diversity: Lower sampling_temp, run multiple seeds
162
163	### Error interpretation
164
165	\| Error \| Cause \| Fix \|
166	\|-------\|-------\|-----\|
167	\| `RuntimeError: CUDA out

People who liked this also viewed

6 picks

Type