boltzgen

Name: boltzgen
Author: adaptyvbio

32installs20forks

How to install boltzgen?

adaptyvbio/protein-design-skills/boltzgen

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

Installs into the current project.

Summary

All-atom protein design using BoltzGen diffusion model. Use this skill when: (1) Need side-chain aware design from the start, (2) Designing around small molecules or ligands, (3) Want all-atom diffusion (not just backbone), (4) Require precise binding geometries, (5) Using YAML-b

Files · 1

View on GitHub

SKILL.md

1	# BoltzGen All-Atom Design
2
3	## Prerequisites
4
5	\| Requirement \| Minimum \| Recommended \|
6	\|-------------\|---------\|-------------\|
7	\| Python \| 3.11+ \| 3.12 \|
8	\| CUDA \| 12.0+ \| 12.1+ \|
9	\| GPU VRAM \| 24GB \| 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	# Clone biomodals
19	git clone https://github.com/hgbrian/biomodals && cd biomodals
20
21	# Run BoltzGen (requires YAML config file)
22	modal run modal_boltzgen.py \
23	--input-yaml binder_config.yaml \
24	--protocol protein-anything \
25	--num-designs 50
26
27	# With custom GPU
28	GPU=L40S modal run modal_boltzgen.py \
29	--input-yaml binder_config.yaml \
30	--protocol protein-anything \
31	--num-designs 100
32	```
33
34	GPU: L40S (48GB) recommended \| Timeout: 120min default
35
36	Available protocols: `protein-anything`, `peptide-anything`, `protein-small_molecule`, `nanobody-anything`, `antibody-anything`
37
38	### Option 2: Local installation
39	```bash
40	git clone https://github.com/HannesStark/boltzgen.git
41	cd boltzgen
42	pip install boltzgen # or: pip install -e .
43
44	# Driven by the boltzgen CLI with a YAML design spec
45	boltzgen run binder_config.yaml \
46	--output out/ \
47	--protocol protein-anything \
48	--num_designs 50
49	```
50
51	The first run downloads model weights (~6GB) to `~/.cache`. Verify a spec with
52	`boltzgen check binder_config.yaml` before a full run.
53
54	GPU: L40S (48GB) \| Time: ~30-60s per design
55
56	## Key parameters (CLI)
57
58	\| Parameter \| Default \| Description \|
59	\|-----------\|---------\|-------------\|
60	\| `--input-yaml` \| required \| Path to YAML design specification \|
61	\| `--protocol` \| `protein-anything` \| Design protocol \|
62	\| `--num-designs` \| 10 \| Number of designs to generate \|
63	\| `--steps` \| all \| Pipeline steps to run (e.g., `design inverse_folding`) \|
64
65	## YAML configuration
66
67	BoltzGen uses an entity-based YAML format where you specify designed proteins and target structures as entities.
68
69	Important notes:
70	- Residue indices use `label_seq_id` (1-indexed), not author residue numbers
71	- File paths are relative to the YAML file location
72	- Target files should be in CIF format (PDB also works but CIF preferred)
73	- Run `boltzgen check config.yaml` to verify your specification before running
74
75	### Basic Binder Config
76	```yaml
77	entities:
78	# Designed protein (variable length 80-140 residues)
79	- protein:
80	id: B
81	sequence: 80..140
82
83	# Target from structure file
84	- file:
85	path: target.cif
86	include:
87	- chain:
88	id: A
89	# Specify binding site residues (optional but recommended)
90	binding_types:
91	- chain:
92	id: A
93	binding: 45,67,89
94	```
95
96	### Binder with Specific Binding Site
97	```yaml
98	entities:
99	- protein:
100	id: G
101	sequence: 60..100
102
103	- file:
104	path: 5cqg.cif
105	include:
106	- chain:
107	id: A
108	binding_types:
109	- chain:
110	id: A
111	binding: 343,344,251
112	structure_groups: "all"
113	```
114
115	### Peptide Design (Cyclic)
116	```yaml
117	entities:
118	- protein:
119	id: S
120	sequence: 10..14C6C3 # With cysteines for disulfide
121
122	- file:
123	path: target.cif
124	include:
125	- chain:
126	id: A
127
128	constraints:
129	- bond:
130	atom1: [S, 11, SG]
131	atom2: [S, 18, SG] # Disulfide bond
132	```
133
134	## Design protocols
135
136	\| Protocol \| Use Case \|
137	\|----------\|----------\|
138	\| `protein-anything` \| Design proteins to bind proteins or peptides \|
139	\| `peptide-anything` \| Design cyclic peptides to bind proteins \|
140	\| `protein-small_molecule` \| Design proteins to bind small molecules \|
141	\| `nanobody-anything` \| Design nanobody CDRs \|
142	\| `antibody-anything` \| Design antibody CDRs \|
143
144	## Output format
145
146	```
147	output/
148	├── sample_0/
149	│ ├── design.cif # All-atom structure (CIF format)
150	│ ├── metrics.json # Confidence scores
151	│ └── sequence.fasta # Sequence
152	├── sample_1/
153	│ └── ...
154	└── summary.csv
155	```
156
157	Note: BoltzGen outputs CIF format. Convert to PDB if needed:
158	```python
159	from Bio.PDB import MMCIFParser, PDBIO
160	parser = MMCIFParser()
161	structure = parser.get_structure("design", "design.cif")
162	io = PDBIO()
163	io.set_structure(structure)
164	io.save("design.pdb")
165	```
166
167	## Sample output
168
169	### Successful run
170	```
171	$ modal run modal_boltzgen.py --input-yam

People who liked this also viewed

6 picks

Type

boltzgen

32installs20forks

View on GitHub

How to install boltzgen?

adaptyvbio/protein-design-skills/boltzgen

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

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# BoltzGen All-Atom Design
2
3	## Prerequisites
4
5	\| Requirement \| Minimum \| Recommended \|
6	\|-------------\|---------\|-------------\|
7	\| Python \| 3.11+ \| 3.12 \|
8	\| CUDA \| 12.0+ \| 12.1+ \|
9	\| GPU VRAM \| 24GB \| 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	# Clone biomodals
19	git clone https://github.com/hgbrian/biomodals && cd biomodals
20
21	# Run BoltzGen (requires YAML config file)
22	modal run modal_boltzgen.py \
23	--input-yaml binder_config.yaml \
24	--protocol protein-anything \
25	--num-designs 50
26
27	# With custom GPU
28	GPU=L40S modal run modal_boltzgen.py \
29	--input-yaml binder_config.yaml \
30	--protocol protein-anything \
31	--num-designs 100
32	```
33
34	GPU: L40S (48GB) recommended \| Timeout: 120min default
35
36	Available protocols: `protein-anything`, `peptide-anything`, `protein-small_molecule`, `nanobody-anything`, `antibody-anything`
37
38	### Option 2: Local installation
39	```bash
40	git clone https://github.com/HannesStark/boltzgen.git
41	cd boltzgen
42	pip install boltzgen # or: pip install -e .
43
44	# Driven by the boltzgen CLI with a YAML design spec
45	boltzgen run binder_config.yaml \
46	--output out/ \
47	--protocol protein-anything \
48	--num_designs 50
49	```
50
51	The first run downloads model weights (~6GB) to `~/.cache`. Verify a spec with
52	`boltzgen check binder_config.yaml` before a full run.
53
54	GPU: L40S (48GB) \| Time: ~30-60s per design
55
56	## Key parameters (CLI)
57
58	\| Parameter \| Default \| Description \|
59	\|-----------\|---------\|-------------\|
60	\| `--input-yaml` \| required \| Path to YAML design specification \|
61	\| `--protocol` \| `protein-anything` \| Design protocol \|
62	\| `--num-designs` \| 10 \| Number of designs to generate \|
63	\| `--steps` \| all \| Pipeline steps to run (e.g., `design inverse_folding`) \|
64
65	## YAML configuration
66
67	BoltzGen uses an entity-based YAML format where you specify designed proteins and target structures as entities.
68
69	Important notes:
70	- Residue indices use `label_seq_id` (1-indexed), not author residue numbers
71	- File paths are relative to the YAML file location
72	- Target files should be in CIF format (PDB also works but CIF preferred)
73	- Run `boltzgen check config.yaml` to verify your specification before running
74
75	### Basic Binder Config
76	```yaml
77	entities:
78	# Designed protein (variable length 80-140 residues)
79	- protein:
80	id: B
81	sequence: 80..140
82
83	# Target from structure file
84	- file:
85	path: target.cif
86	include:
87	- chain:
88	id: A
89	# Specify binding site residues (optional but recommended)
90	binding_types:
91	- chain:
92	id: A
93	binding: 45,67,89
94	```
95
96	### Binder with Specific Binding Site
97	```yaml
98	entities:
99	- protein:
100	id: G
101	sequence: 60..100
102
103	- file:
104	path: 5cqg.cif
105	include:
106	- chain:
107	id: A
108	binding_types:
109	- chain:
110	id: A
111	binding: 343,344,251
112	structure_groups: "all"
113	```
114
115	### Peptide Design (Cyclic)
116	```yaml
117	entities:
118	- protein:
119	id: S
120	sequence: 10..14C6C3 # With cysteines for disulfide
121
122	- file:
123	path: target.cif
124	include:
125	- chain:
126	id: A
127
128	constraints:
129	- bond:
130	atom1: [S, 11, SG]
131	atom2: [S, 18, SG] # Disulfide bond
132	```
133
134	## Design protocols
135
136	\| Protocol \| Use Case \|
137	\|----------\|----------\|
138	\| `protein-anything` \| Design proteins to bind proteins or peptides \|
139	\| `peptide-anything` \| Design cyclic peptides to bind proteins \|
140	\| `protein-small_molecule` \| Design proteins to bind small molecules \|
141	\| `nanobody-anything` \| Design nanobody CDRs \|
142	\| `antibody-anything` \| Design antibody CDRs \|
143
144	## Output format
145
146	```
147	output/
148	├── sample_0/
149	│ ├── design.cif # All-atom structure (CIF format)
150	│ ├── metrics.json # Confidence scores
151	│ └── sequence.fasta # Sequence
152	├── sample_1/
153	│ └── ...
154	└── summary.csv
155	```
156
157	Note: BoltzGen outputs CIF format. Convert to PDB if needed:
158	```python
159	from Bio.PDB import MMCIFParser, PDBIO
160	parser = MMCIFParser()
161	structure = parser.get_structure("design", "design.cif")
162	io = PDBIO()
163	io.set_structure(structure)
164	io.save("design.pdb")
165	```
166
167	## Sample output
168
169	### Successful run
170	```
171	$ modal run modal_boltzgen.py --input-yam

People who liked this also viewed

6 picks

Type