solublempnn

Name: solublempnn
Author: adaptyvbio

29installs20forks

How to install solublempnn?

adaptyvbio/protein-design-skills/solublempnn

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

Installs into the current project.

Summary

Solubility-optimized protein sequence design using SolubleMPNN. Use this skill when: (1) Designing for E. coli expression, (2) Optimizing solubility of designed proteins, (3) Reducing aggregation propensity, (4) Need high-yield expression, (5) Avoiding inclusion body formation. F

Files · 1

View on GitHub

SKILL.md

1	# SolubleMPNN Solubility-Optimized Design
2
3	## Prerequisites
4
5	\| Requirement \| Minimum \| Recommended \|
6	\|-------------\|---------\|-------------\|
7	\| Python \| 3.8+ \| 3.10 \|
8	\| CUDA \| 11.0+ \| 11.7+ \|
9	\| GPU VRAM \| 8GB \| 16GB (T4) \|
10	\| RAM \| 8GB \| 16GB \|
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	SolubleMPNN is the soluble model type within the LigandMPNN wrapper:
18	```bash
19	cd biomodals
20	modal run modal_ligandmpnn.py \
21	--input-pdb backbone.pdb \
22	--params-str "--model_type soluble_mpnn --number_of_batches 16 --temperature 0.1"
23	```
24
25	GPU: A10G default \| Timeout: 900s default
26
27	### Option 2: Local installation
28	```bash
29	git clone https://github.com/dauparas/ProteinMPNN.git
30	cd ProteinMPNN
31
32	# The soluble weights are selected with --use_soluble_model, not a model name
33	python protein_mpnn_run.py \
34	--pdb_path backbone.pdb \
35	--out_folder output/ \
36	--num_seq_per_target 16 \
37	--sampling_temp "0.1" \
38	--use_soluble_model
39	```
40
41	## Key parameters
42
43	\| Parameter \| Default \| Description \|
44	\|-----------\|---------\|-------------\|
45	\| `--pdb_path` \| required \| Input structure \|
46	\| `--use_soluble_model` \| off \| Use the solubility-trained weights \|
47	\| `--num_seq_per_target` \| 1 \| Sequences per structure \|
48	\| `--sampling_temp` \| "0.1" \| Temperature (string) \|
49	\| `--model_name` \| v_48_020 \| Noise level (0.20 A); orthogonal to solubility \|
50
51	## Model weights
52
53	`--model_name` sets the training-noise level (v_48_002 = 0.02 A, v_48_010 = 0.10 A,
54	v_48_020 = 0.20 A), not a solubility tier. Solubility is a separate weight set chosen
55	with `--use_soluble_model`, available for v_48_010 and v_48_020. Higher noise gives
56	more sequence diversity.
57
58	## Output format
59
60	```
61	output/
62	├── seqs/backbone.fa
63	└── backbone_pdb/backbone_0001.pdb
64	```
65
66	## Sample output
67
68	### Successful run
69	```
70	$ python protein_mpnn_run.py --pdb_path backbone.pdb --use_soluble_model --num_seq_per_target 8
71	Loading soluble model weights (v_48_020)...
72	Designing sequences for backbone.pdb
73	Generated 8 sequences in 2.1 seconds
74
75	output/seqs/backbone.fa:
76	>backbone_0001, score=1.31, global_score=1.24, seq_recovery=0.78
77	MKTAYIAKQRQISFVKSHFSRQLE...
78	>backbone_0002, score=1.28, global_score=1.21, seq_recovery=0.81
79	MKTAYIAKQRQISFVKSQFSRQLD...
80	```
81
82	What good output looks like:
83	- Score: 1.0-2.0 (lower = more confident)
84	- Reduced hydrophobic patches compared to standard MPNN
85	- Improved charge distribution
86
87	## Decision tree
88
89	```
90	Should I use SolubleMPNN?
91	│
92	├─ What expression system?
93	│ ├─ E. coli → SolubleMPNN ✓
94	│ ├─ Mammalian → ProteinMPNN (PTMs matter more)
95	│ └─ Yeast → Either
96	│
97	├─ History of expression problems?
98	│ ├─ Yes, aggregation → SolubleMPNN ✓
99	│ ├─ Yes, low yield → SolubleMPNN ✓
100	│ └─ No → ProteinMPNN is fine
101	│
102	├─ What's in the binding site?
103	│ ├─ Small molecule / ligand → Use LigandMPNN
104	│ └─ Nothing / protein only → SolubleMPNN ✓
105	│
106	└─ Optimizing for expression?
107	└─ Add --use_soluble_model to ProteinMPNN
108	```
109
110	## Typical performance
111
112	\| Campaign Size \| Time (T4) \| Cost (Modal) \| Notes \|
113	\|---------------\|-----------\|--------------\|-------\|
114	\| 100 backbones × 8 seq \| 15-20 min \| ~$2 \| Standard \|
115	\| 500 backbones × 8 seq \| 1-1.5h \| ~$8 \| Large campaign \|
116
117	Expected improvement: +15-30% solubility score vs standard ProteinMPNN.
118
119	---
120
121	## Verify
122
123	```bash
124	grep -c "^>" output/seqs/*.fa # Should match backbone_count × num_seq_per_target
125	```
126
127	---
128
129	## Troubleshooting
130
131	Still insoluble: Confirm `--use_soluble_model` is set; redesign more positions or add explicit hydrophobic-residue bias
132	Low diversity: Increase temperature to 0.2
133	Poor folding: Use standard ProteinMPNN and optimize later
134
135	### Error interpretation
136
137	\| Error \| Cause \| Fix \|
138	\|-------\|-------\|-----\|
139	\| `RuntimeError: CUDA out of memory` \| Long protein or large batch \| Reduce batch_size \|
140	\| `FileNotFoundError: v_48_020` \| Missing model weights \| Download soluble weights \|
141
142	---
143
144	Next: Structure prediction for validation → `protein-qc` for filtering.

People who liked this also viewed

6 picks

Type

solublempnn

29installs20forks

View on GitHub

How to install solublempnn?

adaptyvbio/protein-design-skills/solublempnn

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

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# SolubleMPNN Solubility-Optimized Design
2
3	## Prerequisites
4
5	\| Requirement \| Minimum \| Recommended \|
6	\|-------------\|---------\|-------------\|
7	\| Python \| 3.8+ \| 3.10 \|
8	\| CUDA \| 11.0+ \| 11.7+ \|
9	\| GPU VRAM \| 8GB \| 16GB (T4) \|
10	\| RAM \| 8GB \| 16GB \|
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	SolubleMPNN is the soluble model type within the LigandMPNN wrapper:
18	```bash
19	cd biomodals
20	modal run modal_ligandmpnn.py \
21	--input-pdb backbone.pdb \
22	--params-str "--model_type soluble_mpnn --number_of_batches 16 --temperature 0.1"
23	```
24
25	GPU: A10G default \| Timeout: 900s default
26
27	### Option 2: Local installation
28	```bash
29	git clone https://github.com/dauparas/ProteinMPNN.git
30	cd ProteinMPNN
31
32	# The soluble weights are selected with --use_soluble_model, not a model name
33	python protein_mpnn_run.py \
34	--pdb_path backbone.pdb \
35	--out_folder output/ \
36	--num_seq_per_target 16 \
37	--sampling_temp "0.1" \
38	--use_soluble_model
39	```
40
41	## Key parameters
42
43	\| Parameter \| Default \| Description \|
44	\|-----------\|---------\|-------------\|
45	\| `--pdb_path` \| required \| Input structure \|
46	\| `--use_soluble_model` \| off \| Use the solubility-trained weights \|
47	\| `--num_seq_per_target` \| 1 \| Sequences per structure \|
48	\| `--sampling_temp` \| "0.1" \| Temperature (string) \|
49	\| `--model_name` \| v_48_020 \| Noise level (0.20 A); orthogonal to solubility \|
50
51	## Model weights
52
53	`--model_name` sets the training-noise level (v_48_002 = 0.02 A, v_48_010 = 0.10 A,
54	v_48_020 = 0.20 A), not a solubility tier. Solubility is a separate weight set chosen
55	with `--use_soluble_model`, available for v_48_010 and v_48_020. Higher noise gives
56	more sequence diversity.
57
58	## Output format
59
60	```
61	output/
62	├── seqs/backbone.fa
63	└── backbone_pdb/backbone_0001.pdb
64	```
65
66	## Sample output
67
68	### Successful run
69	```
70	$ python protein_mpnn_run.py --pdb_path backbone.pdb --use_soluble_model --num_seq_per_target 8
71	Loading soluble model weights (v_48_020)...
72	Designing sequences for backbone.pdb
73	Generated 8 sequences in 2.1 seconds
74
75	output/seqs/backbone.fa:
76	>backbone_0001, score=1.31, global_score=1.24, seq_recovery=0.78
77	MKTAYIAKQRQISFVKSHFSRQLE...
78	>backbone_0002, score=1.28, global_score=1.21, seq_recovery=0.81
79	MKTAYIAKQRQISFVKSQFSRQLD...
80	```
81
82	What good output looks like:
83	- Score: 1.0-2.0 (lower = more confident)
84	- Reduced hydrophobic patches compared to standard MPNN
85	- Improved charge distribution
86
87	## Decision tree
88
89	```
90	Should I use SolubleMPNN?
91	│
92	├─ What expression system?
93	│ ├─ E. coli → SolubleMPNN ✓
94	│ ├─ Mammalian → ProteinMPNN (PTMs matter more)
95	│ └─ Yeast → Either
96	│
97	├─ History of expression problems?
98	│ ├─ Yes, aggregation → SolubleMPNN ✓
99	│ ├─ Yes, low yield → SolubleMPNN ✓
100	│ └─ No → ProteinMPNN is fine
101	│
102	├─ What's in the binding site?
103	│ ├─ Small molecule / ligand → Use LigandMPNN
104	│ └─ Nothing / protein only → SolubleMPNN ✓
105	│
106	└─ Optimizing for expression?
107	└─ Add --use_soluble_model to ProteinMPNN
108	```
109
110	## Typical performance
111
112	\| Campaign Size \| Time (T4) \| Cost (Modal) \| Notes \|
113	\|---------------\|-----------\|--------------\|-------\|
114	\| 100 backbones × 8 seq \| 15-20 min \| ~$2 \| Standard \|
115	\| 500 backbones × 8 seq \| 1-1.5h \| ~$8 \| Large campaign \|
116
117	Expected improvement: +15-30% solubility score vs standard ProteinMPNN.
118
119	---
120
121	## Verify
122
123	```bash
124	grep -c "^>" output/seqs/*.fa # Should match backbone_count × num_seq_per_target
125	```
126
127	---
128
129	## Troubleshooting
130
131	Still insoluble: Confirm `--use_soluble_model` is set; redesign more positions or add explicit hydrophobic-residue bias
132	Low diversity: Increase temperature to 0.2
133	Poor folding: Use standard ProteinMPNN and optimize later
134
135	### Error interpretation
136
137	\| Error \| Cause \| Fix \|
138	\|-------\|-------\|-----\|
139	\| `RuntimeError: CUDA out of memory` \| Long protein or large batch \| Reduce batch_size \|
140	\| `FileNotFoundError: v_48_020` \| Missing model weights \| Download soluble weights \|
141
142	---
143
144	Next: Structure prediction for validation → `protein-qc` for filtering.

People who liked this also viewed

6 picks

Type