alphafold

Name: alphafold
Author: adaptyvbio

44installs20forks

How to install alphafold?

adaptyvbio/protein-design-skills/alphafold

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

Installs into the current project.

Summary

Validate protein designs using AlphaFold2 structure prediction. Use this skill when: (1) Validating designed sequences fold correctly, (2) Predicting binder-target complex structures, (3) Calculating confidence metrics (pLDDT, pTM, ipTM), (4) Self-consistency validation of design

Files · 1

View on GitHub

SKILL.md

1	# AlphaFold2 Structure Validation
2
3	## Prerequisites
4
5	\| Requirement \| Minimum \| Recommended \|
6	\|-------------\|---------\|-------------\|
7	\| Python \| 3.8+ \| 3.10 \|
8	\| CUDA \| 11.0+ \| 12.0+ \|
9	\| GPU VRAM \| 32GB \| 40GB (A100) \|
10	\| RAM \| 32GB \| 64GB \|
11	\| Disk \| 100GB \| 500GB (for databases) \|
12
13	## How to run
14
15	> First time? See [Getting started](../../docs/getting-started.md) to set up Modal and biomodals.
16
17	### Option 1: Modal (AlphaFold-Multimer)
18	```bash
19	cd biomodals
20	modal run modal_alphafold.py \
21	--input-fasta sequences.fasta \
22	--out-dir output/
23	```
24
25	GPU: A100 (40GB) \| Timeout: 3600s default
26
27	### Option 2: Local installation
28	```bash
29	git clone https://github.com/google-deepmind/alphafold.git
30	cd alphafold
31
32	python run_alphafold.py \
33	--fasta_paths=query.fasta \
34	--output_dir=output/ \
35	--model_preset=monomer \
36	--max_template_date=2026-01-01
37	```
38
39	### Option 3: ESMFold2 (fast single-sequence)
40	```bash
41	printf '>protein\|A\nMKTAYIAKQRQISFVK...\n' > seq.faa
42	uv run --with modal modal run modal_esmfold2.py --input-faa seq.faa
43	```
44
45	## Key parameters
46
47	\| Parameter \| Default \| Options \| Description \|
48	\|-----------\|---------\|---------\|-------------\|
49	\| `--model_preset` \| monomer \| monomer/multimer \| Model type \|
50	\| `--num_recycle` \| 3 \| 1-20 \| Recycling iterations \|
51	\| `--max_template_date` \| - \| YYYY-MM-DD \| Template cutoff \|
52	\| `--use_templates` \| True \| True/False \| Use template search \|
53
54	## Output format
55
56	```
57	output/
58	├── ranked_0.pdb # Best model
59	├── ranked_1.pdb # Second best
60	├── ranking_debug.json # Confidence scores
61	├── result_model_1.pkl # Full results
62	├── msas/ # MSA files
63	└── features.pkl # Input features
64	```
65
66	### Extracting metrics
67	```python
68	import pickle
69
70	with open('result_model_1.pkl', 'rb') as f:
71	result = pickle.load(f)
72
73	plddt = result['plddt']
74	ptm = result['ptm']
75	iptm = result.get('iptm', None) # Multimer only
76	pae = result['predicted_aligned_error']
77	```
78
79	## Sample output
80
81	### Successful run
82	```
83	$ python run_alphafold.py --fasta_paths complex.fasta --model_preset multimer
84	[INFO] Running MSA search...
85	[INFO] Running model 1/5...
86	[INFO] Running model 5/5...
87	[INFO] Relaxing structures...
88
89	Results:
90	ranked_0.pdb:
91	pLDDT: 87.3 (mean)
92	pTM: 0.78
93	ipTM: 0.62
94	PAE (interface): 8.5
95
96	Saved to output/
97	```
98
99	What good output looks like:
100	- pLDDT: > 85 (mean, on 0-100 scale) or > 0.85 (normalized)
101	- pTM: > 0.70
102	- ipTM: > 0.50 for complexes
103	- PAE_interface: < 10
104
105	## Decision tree
106
107	```
108	Should I use AlphaFold?
109	│
110	├─ What are you predicting?
111	│ ├─ Single protein → ESMFold (faster)
112	│ ├─ Protein-protein complex → AlphaFold/ColabFold ✓
113	│ ├─ Protein + ligand → Chai or Boltz
114	│ └─ Batch of sequences → ColabFold ✓
115	│
116	├─ What do you need?
117	│ ├─ Highest accuracy → AlphaFold/ColabFold ✓
118	│ ├─ Fast screening → ESMFold
119	│ └─ MSA-free prediction → Chai or ESMFold
120	│
121	└─ Which AF2 option?
122	├─ Local installation → Full control, slow setup
123	├─ ColabFold → Easier, MSA server
124	└─ Modal → Recommended for batch
125	```
126
127	## Typical performance
128
129	\| Campaign Size \| Time (A100) \| Cost (Modal) \| Notes \|
130	\|---------------\|-------------\|--------------\|-------\|
131	\| 100 complexes \| 1-2h \| ~$8 \| With MSA server \|
132	\| 500 complexes \| 5-10h \| ~$40 \| Standard campaign \|
133	\| 1000 complexes \| 10-20h \| ~$80 \| Large campaign \|
134
135	Per-complex: ~30-60s with MSA server.
136
137	---
138
139	## Verify
140
141	```bash
142	find output -name "ranked_0.pdb" \| wc -l # Should match input count
143	```
144
145	---
146
147	## Troubleshooting
148
149	Low pLDDT regions: May indicate disorder or poor design
150	Low ipTM: Interface not confident, check hotspots
151	High PAE off-diagonal: Chains may not interact
152	OOM errors: Use ColabFold with MSA server instead
153
154	### Error interpretation
155
156	\| Error \| Cause \| Fix \|
157	\|-------\|-------\|-----\|
158	\| `RuntimeError: CUDA out of memory` \| Sequence too long \| Use A100 or split prediction \|
159	\| `KeyError: 'iptm'` \| Running monomer on complex \| Use multimer preset \|
160	\| `FileNotFoundError: database` \| Missing MSA databases \| Use ColabFold MSA server \|
161	\| `TimeoutError` \| MSA search slow \| Reduce num_recycles \|
162
163	---
164
165	Next: `protein-qc` for filtering and ranking.

People who liked this also viewed

6 picks

Type

alphafold

44installs20forks

View on GitHub

How to install alphafold?

adaptyvbio/protein-design-skills/alphafold

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

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# AlphaFold2 Structure Validation
2
3	## Prerequisites
4
5	\| Requirement \| Minimum \| Recommended \|
6	\|-------------\|---------\|-------------\|
7	\| Python \| 3.8+ \| 3.10 \|
8	\| CUDA \| 11.0+ \| 12.0+ \|
9	\| GPU VRAM \| 32GB \| 40GB (A100) \|
10	\| RAM \| 32GB \| 64GB \|
11	\| Disk \| 100GB \| 500GB (for databases) \|
12
13	## How to run
14
15	> First time? See [Getting started](../../docs/getting-started.md) to set up Modal and biomodals.
16
17	### Option 1: Modal (AlphaFold-Multimer)
18	```bash
19	cd biomodals
20	modal run modal_alphafold.py \
21	--input-fasta sequences.fasta \
22	--out-dir output/
23	```
24
25	GPU: A100 (40GB) \| Timeout: 3600s default
26
27	### Option 2: Local installation
28	```bash
29	git clone https://github.com/google-deepmind/alphafold.git
30	cd alphafold
31
32	python run_alphafold.py \
33	--fasta_paths=query.fasta \
34	--output_dir=output/ \
35	--model_preset=monomer \
36	--max_template_date=2026-01-01
37	```
38
39	### Option 3: ESMFold2 (fast single-sequence)
40	```bash
41	printf '>protein\|A\nMKTAYIAKQRQISFVK...\n' > seq.faa
42	uv run --with modal modal run modal_esmfold2.py --input-faa seq.faa
43	```
44
45	## Key parameters
46
47	\| Parameter \| Default \| Options \| Description \|
48	\|-----------\|---------\|---------\|-------------\|
49	\| `--model_preset` \| monomer \| monomer/multimer \| Model type \|
50	\| `--num_recycle` \| 3 \| 1-20 \| Recycling iterations \|
51	\| `--max_template_date` \| - \| YYYY-MM-DD \| Template cutoff \|
52	\| `--use_templates` \| True \| True/False \| Use template search \|
53
54	## Output format
55
56	```
57	output/
58	├── ranked_0.pdb # Best model
59	├── ranked_1.pdb # Second best
60	├── ranking_debug.json # Confidence scores
61	├── result_model_1.pkl # Full results
62	├── msas/ # MSA files
63	└── features.pkl # Input features
64	```
65
66	### Extracting metrics
67	```python
68	import pickle
69
70	with open('result_model_1.pkl', 'rb') as f:
71	result = pickle.load(f)
72
73	plddt = result['plddt']
74	ptm = result['ptm']
75	iptm = result.get('iptm', None) # Multimer only
76	pae = result['predicted_aligned_error']
77	```
78
79	## Sample output
80
81	### Successful run
82	```
83	$ python run_alphafold.py --fasta_paths complex.fasta --model_preset multimer
84	[INFO] Running MSA search...
85	[INFO] Running model 1/5...
86	[INFO] Running model 5/5...
87	[INFO] Relaxing structures...
88
89	Results:
90	ranked_0.pdb:
91	pLDDT: 87.3 (mean)
92	pTM: 0.78
93	ipTM: 0.62
94	PAE (interface): 8.5
95
96	Saved to output/
97	```
98
99	What good output looks like:
100	- pLDDT: > 85 (mean, on 0-100 scale) or > 0.85 (normalized)
101	- pTM: > 0.70
102	- ipTM: > 0.50 for complexes
103	- PAE_interface: < 10
104
105	## Decision tree
106
107	```
108	Should I use AlphaFold?
109	│
110	├─ What are you predicting?
111	│ ├─ Single protein → ESMFold (faster)
112	│ ├─ Protein-protein complex → AlphaFold/ColabFold ✓
113	│ ├─ Protein + ligand → Chai or Boltz
114	│ └─ Batch of sequences → ColabFold ✓
115	│
116	├─ What do you need?
117	│ ├─ Highest accuracy → AlphaFold/ColabFold ✓
118	│ ├─ Fast screening → ESMFold
119	│ └─ MSA-free prediction → Chai or ESMFold
120	│
121	└─ Which AF2 option?
122	├─ Local installation → Full control, slow setup
123	├─ ColabFold → Easier, MSA server
124	└─ Modal → Recommended for batch
125	```
126
127	## Typical performance
128
129	\| Campaign Size \| Time (A100) \| Cost (Modal) \| Notes \|
130	\|---------------\|-------------\|--------------\|-------\|
131	\| 100 complexes \| 1-2h \| ~$8 \| With MSA server \|
132	\| 500 complexes \| 5-10h \| ~$40 \| Standard campaign \|
133	\| 1000 complexes \| 10-20h \| ~$80 \| Large campaign \|
134
135	Per-complex: ~30-60s with MSA server.
136
137	---
138
139	## Verify
140
141	```bash
142	find output -name "ranked_0.pdb" \| wc -l # Should match input count
143	```
144
145	---
146
147	## Troubleshooting
148
149	Low pLDDT regions: May indicate disorder or poor design
150	Low ipTM: Interface not confident, check hotspots
151	High PAE off-diagonal: Chains may not interact
152	OOM errors: Use ColabFold with MSA server instead
153
154	### Error interpretation
155
156	\| Error \| Cause \| Fix \|
157	\|-------\|-------\|-----\|
158	\| `RuntimeError: CUDA out of memory` \| Sequence too long \| Use A100 or split prediction \|
159	\| `KeyError: 'iptm'` \| Running monomer on complex \| Use multimer preset \|
160	\| `FileNotFoundError: database` \| Missing MSA databases \| Use ColabFold MSA server \|
161	\| `TimeoutError` \| MSA search slow \| Reduce num_recycles \|
162
163	---
164
165	Next: `protein-qc` for filtering and ranking.

People who liked this also viewed

6 picks

Type