chai

Name: chai
Author: adaptyvbio

37installs20forks

How to install chai?

adaptyvbio/protein-design-skills/chai

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

Installs into the current project.

Summary

Structure prediction using Chai-1, a foundation model for molecular structure. Use this skill when: (1) Predicting protein-protein complex structures, (2) Validating designed binders, (3) Predicting protein-ligand complexes, (4) Using the Chai API for high-throughput prediction,

Files · 1

View on GitHub

SKILL.md

1	# Chai-1 Structure Prediction
2
3	## Prerequisites
4
5	\| Requirement \| Minimum \| Recommended \|
6	\|-------------\|---------\|-------------\|
7	\| Python \| 3.10+ \| 3.11 \|
8	\| CUDA \| 12.0+ \| 12.1+ \|
9	\| GPU VRAM \| 24GB \| 40GB (A100) \|
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
17	```bash
18	cd biomodals
19	modal run modal_chai1.py \
20	--input-faa complex.fasta \
21	--out-dir predictions/
22	```
23
24	GPU: A100 (40GB) \| Timeout: 30min default
25
26	### Option 2: Chai API (recommended)
27	```bash
28	pip install chai_lab
29
30	python -c "
31	import chai_lab
32	from chai_lab.chai1 import run_inference
33
34	# Run prediction
35	run_inference(
36	fasta_file='complex.fasta',
37	output_dir='predictions/',
38	num_trunk_recycles=3
39	)
40	"
41	```
42
43	### Option 3: Local installation
44	```bash
45	git clone https://github.com/chaidiscovery/chai-lab.git
46	cd chai-lab
47	pip install -e .
48
49	chai-lab predict \
50	--fasta complex.fasta \
51	--output predictions/
52	```
53
54	## FASTA Format
55
56	### Protein complex
57	```
58	>binder
59	MKTAYIAKQRQISFVKSHFSRQLE...
60	>target
61	MVLSPADKTNVKAAWGKVGAHAGE...
62	```
63
64	### Protein + ligand
65	```
66	>protein
67	MKTAYIAKQRQISFVKSHFSRQLE...
68	>ligand\|smiles
69	CCO
70	```
71
72	### Protein + DNA/RNA
73	```
74	>protein
75	MKTAYIAKQRQISFVKSHFSRQLE...
76	>dna
77	ATCGATCGATCG
78	```
79
80	## Key parameters
81
82	\| Parameter \| Default \| Range \| Description \|
83	\|-----------\|---------\|-------\|-------------\|
84	\| `num_trunk_recycles` \| 3 \| 1-10 \| Recycles (more = better) \|
85	\| `num_diffn_timesteps` \| 200 \| 50-500 \| Diffusion steps \|
86	\| `seed` \| 0 \| int \| Random seed \|
87
88	## Output format
89
90	```
91	predictions/
92	├── pred.model_idx_0.cif # Best model (CIF format)
93	├── pred.model_idx_1.cif # Second model
94	├── scores.json # Confidence scores
95	├── pae.npy # PAE matrix
96	└── plddt.npy # pLDDT values
97	```
98
99	Note: Chai-1 outputs CIF format. Convert to PDB if needed:
100	```python
101	from Bio.PDB import MMCIFParser, PDBIO
102	parser = MMCIFParser()
103	structure = parser.get_structure("pred", "pred.model_idx_0.cif")
104	io = PDBIO()
105	io.set_structure(structure)
106	io.save("pred.model_idx_0.pdb")
107	```
108
109	### Extracting metrics
110	```python
111	import numpy as np
112	import json
113
114	# Load scores
115	with open('predictions/scores.json') as f:
116	scores = json.load(f)
117
118	plddt = np.load('predictions/plddt.npy')
119	pae = np.load('predictions/pae.npy')
120
121	print(f"pLDDT: {plddt.mean():.3f}")
122	print(f"pTM: {scores['ptm']:.3f}")
123	print(f"ipTM: {scores.get('iptm', 'N/A')}")
124	```
125
126	## Use cases
127
128	### Binder validation
129	```python
130	# Predict complex with Chai
131	chai-lab predict --fasta binder_target.fasta --output val/
132
133	# Check ipTM > 0.5
134	scores = json.load(open('val/scores.json'))
135	if scores['iptm'] > 0.5:
136	print("Design passes validation")
137	```
138
139	### Protein-ligand complex
140	```python
141	# FASTA with SMILES
142	fasta = """
143	>protein
144	MKTA...
145	>ligand\|smiles
146	CCO
147	"""
148
149	# Chai handles both protein and small molecules
150	```
151
152	### Batch prediction
153	```bash
154	# Multiple sequences
155	for fasta in sequences/*.fasta; do
156	chai-lab predict \
157	--fasta "$fasta" \
158	--output "predictions/$(basename $fasta .fasta)"
159	done
160	```
161
162	## Comparison with AF2
163
164	\| Aspect \| Chai-1 \| AlphaFold2 \|
165	\|--------\|--------\|------------\|
166	\| MSA required \| No \| Yes \|
167	\| Small molecules \| Yes \| No \|
168	\| DNA/RNA \| Yes \| Limited \|
169	\| Speed \| Faster \| Slower \|
170	\| Accuracy \| Comparable \| Reference \|
171
172	## Sample output
173
174	### Successful run
175	```
176	$ chai-lab predict --fasta complex.fasta --output predictions/
177	[INFO] Loading Chai-1 model...
178	[INFO] Running inference...
179	[INFO] Saved 5 models to predictions/
180
181	predictions/scores.json:
182	{
183	"ptm": 0.82,
184	"iptm": 0.71,
185	"ranking_score": 0.76
186	}
187	```
188
189	What good output looks like:
190	- pTM: > 0.7 (confident global structure)
191	- ipTM: > 0.5 (confident interface, > 0.7 for high confidence)
192	- CIF files with reasonable atom positions
193
194	## Decision tree
195
196	```
197	Should I use Chai?
198	│
199	├─ What are you predicting?
200	│ ├─ Protein-protein complex → Chai ✓ or ColabFold
201	│ ├─ Protein + small molecule → Chai ✓
202	│ ├─ Protein + DNA/RNA → Chai ✓
203	│ └─ Single protein only → Use ESMFold (faster)
204	│
205	├─ Need MSA?
206	│ ├─ No / want speed → Chai ✓
207	│ └─ Yes / want accuracy → ColabFold
208	│
209	└─ Priority?
210	├─ Highest accuracy → ColabFold with MSA
211	├─ Speed / no MSA → Chai ✓
212	└─ Ligand binding → Chai ✓
213	```
214
215	## Typical performance
216
217	\| Campaign Size \| Time (A100) \| Cost (Modal) \| Notes \|
218	\|---------------\|-------------\|--------------\|-------\|

People who liked this also viewed

6 picks

Type

chai

37installs20forks

View on GitHub

How to install chai?

adaptyvbio/protein-design-skills/chai

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

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Chai-1 Structure Prediction
2
3	## Prerequisites
4
5	\| Requirement \| Minimum \| Recommended \|
6	\|-------------\|---------\|-------------\|
7	\| Python \| 3.10+ \| 3.11 \|
8	\| CUDA \| 12.0+ \| 12.1+ \|
9	\| GPU VRAM \| 24GB \| 40GB (A100) \|
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
17	```bash
18	cd biomodals
19	modal run modal_chai1.py \
20	--input-faa complex.fasta \
21	--out-dir predictions/
22	```
23
24	GPU: A100 (40GB) \| Timeout: 30min default
25
26	### Option 2: Chai API (recommended)
27	```bash
28	pip install chai_lab
29
30	python -c "
31	import chai_lab
32	from chai_lab.chai1 import run_inference
33
34	# Run prediction
35	run_inference(
36	fasta_file='complex.fasta',
37	output_dir='predictions/',
38	num_trunk_recycles=3
39	)
40	"
41	```
42
43	### Option 3: Local installation
44	```bash
45	git clone https://github.com/chaidiscovery/chai-lab.git
46	cd chai-lab
47	pip install -e .
48
49	chai-lab predict \
50	--fasta complex.fasta \
51	--output predictions/
52	```
53
54	## FASTA Format
55
56	### Protein complex
57	```
58	>binder
59	MKTAYIAKQRQISFVKSHFSRQLE...
60	>target
61	MVLSPADKTNVKAAWGKVGAHAGE...
62	```
63
64	### Protein + ligand
65	```
66	>protein
67	MKTAYIAKQRQISFVKSHFSRQLE...
68	>ligand\|smiles
69	CCO
70	```
71
72	### Protein + DNA/RNA
73	```
74	>protein
75	MKTAYIAKQRQISFVKSHFSRQLE...
76	>dna
77	ATCGATCGATCG
78	```
79
80	## Key parameters
81
82	\| Parameter \| Default \| Range \| Description \|
83	\|-----------\|---------\|-------\|-------------\|
84	\| `num_trunk_recycles` \| 3 \| 1-10 \| Recycles (more = better) \|
85	\| `num_diffn_timesteps` \| 200 \| 50-500 \| Diffusion steps \|
86	\| `seed` \| 0 \| int \| Random seed \|
87
88	## Output format
89
90	```
91	predictions/
92	├── pred.model_idx_0.cif # Best model (CIF format)
93	├── pred.model_idx_1.cif # Second model
94	├── scores.json # Confidence scores
95	├── pae.npy # PAE matrix
96	└── plddt.npy # pLDDT values
97	```
98
99	Note: Chai-1 outputs CIF format. Convert to PDB if needed:
100	```python
101	from Bio.PDB import MMCIFParser, PDBIO
102	parser = MMCIFParser()
103	structure = parser.get_structure("pred", "pred.model_idx_0.cif")
104	io = PDBIO()
105	io.set_structure(structure)
106	io.save("pred.model_idx_0.pdb")
107	```
108
109	### Extracting metrics
110	```python
111	import numpy as np
112	import json
113
114	# Load scores
115	with open('predictions/scores.json') as f:
116	scores = json.load(f)
117
118	plddt = np.load('predictions/plddt.npy')
119	pae = np.load('predictions/pae.npy')
120
121	print(f"pLDDT: {plddt.mean():.3f}")
122	print(f"pTM: {scores['ptm']:.3f}")
123	print(f"ipTM: {scores.get('iptm', 'N/A')}")
124	```
125
126	## Use cases
127
128	### Binder validation
129	```python
130	# Predict complex with Chai
131	chai-lab predict --fasta binder_target.fasta --output val/
132
133	# Check ipTM > 0.5
134	scores = json.load(open('val/scores.json'))
135	if scores['iptm'] > 0.5:
136	print("Design passes validation")
137	```
138
139	### Protein-ligand complex
140	```python
141	# FASTA with SMILES
142	fasta = """
143	>protein
144	MKTA...
145	>ligand\|smiles
146	CCO
147	"""
148
149	# Chai handles both protein and small molecules
150	```
151
152	### Batch prediction
153	```bash
154	# Multiple sequences
155	for fasta in sequences/*.fasta; do
156	chai-lab predict \
157	--fasta "$fasta" \
158	--output "predictions/$(basename $fasta .fasta)"
159	done
160	```
161
162	## Comparison with AF2
163
164	\| Aspect \| Chai-1 \| AlphaFold2 \|
165	\|--------\|--------\|------------\|
166	\| MSA required \| No \| Yes \|
167	\| Small molecules \| Yes \| No \|
168	\| DNA/RNA \| Yes \| Limited \|
169	\| Speed \| Faster \| Slower \|
170	\| Accuracy \| Comparable \| Reference \|
171
172	## Sample output
173
174	### Successful run
175	```
176	$ chai-lab predict --fasta complex.fasta --output predictions/
177	[INFO] Loading Chai-1 model...
178	[INFO] Running inference...
179	[INFO] Saved 5 models to predictions/
180
181	predictions/scores.json:
182	{
183	"ptm": 0.82,
184	"iptm": 0.71,
185	"ranking_score": 0.76
186	}
187	```
188
189	What good output looks like:
190	- pTM: > 0.7 (confident global structure)
191	- ipTM: > 0.5 (confident interface, > 0.7 for high confidence)
192	- CIF files with reasonable atom positions
193
194	## Decision tree
195
196	```
197	Should I use Chai?
198	│
199	├─ What are you predicting?
200	│ ├─ Protein-protein complex → Chai ✓ or ColabFold
201	│ ├─ Protein + small molecule → Chai ✓
202	│ ├─ Protein + DNA/RNA → Chai ✓
203	│ └─ Single protein only → Use ESMFold (faster)
204	│
205	├─ Need MSA?
206	│ ├─ No / want speed → Chai ✓
207	│ └─ Yes / want accuracy → ColabFold
208	│
209	└─ Priority?
210	├─ Highest accuracy → ColabFold with MSA
211	├─ Speed / no MSA → Chai ✓
212	└─ Ligand binding → Chai ✓
213	```
214
215	## Typical performance
216
217	\| Campaign Size \| Time (A100) \| Cost (Modal) \| Notes \|
218	\|---------------\|-------------\|--------------\|-------\|

People who liked this also viewed

6 picks

Type