ligandmpnn

Name: ligandmpnn
Author: adaptyvbio

30installs20forks

How to install ligandmpnn?

adaptyvbio/protein-design-skills/ligandmpnn

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

Installs into the current project.

Summary

Ligand-aware protein sequence design using LigandMPNN. Use this skill when: (1) Designing sequences around small molecules, (2) Enzyme active site design, (3) Ligand binding pocket optimization, (4) Metal coordination site design, (5) Cofactor binding proteins. For standard prote

Files · 1

View on GitHub

SKILL.md

1	# LigandMPNN Ligand-Aware 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	```bash
18	cd biomodals
19	# modal_ligandmpnn.py takes --input-pdb; LigandMPNN run.py args go in --params-str
20	modal run modal_ligandmpnn.py \
21	--input-pdb protein_ligand.pdb \
22	--params-str "--model_type ligand_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/LigandMPNN.git
30	cd LigandMPNN
31
32	python run.py \
33	--model_type ligand_mpnn \
34	--pdb_path protein_ligand.pdb \
35	--out_folder output/ \
36	--number_of_batches 16 \
37	--temperature 0.1
38	```
39
40	## Key parameters (LigandMPNN run.py)
41
42	\| Parameter \| Default \| Description \|
43	\|-----------\|---------\|-------------\|
44	\| `--pdb_path` \| required \| PDB with ligand \|
45	\| `--model_type` \| `protein_mpnn` \| `ligand_mpnn`, `soluble_mpnn`, etc. \|
46	\| `--temperature` \| 0.1 \| Sampling temperature \|
47	\| `--number_of_batches` \| 1 \| Batches (sequences = batch_size x batches) \|
48	\| `--batch_size` \| 1 \| Sequences per batch \|
49	\| `--ligand_mpnn_use_side_chain_context` \| 0 \| Use ligand side-chain context \|
50
51	## Ligand Specification
52
53	### In PDB File
54	Ligand must be present as HETATM records:
55	```
56	ATOM ...protein atoms...
57	HETATM 1 C1 LIG A 999 x.xxx y.yyy z.zzz 1.00 0.00 C
58	```
59
60	### Supported Ligand Types
61	- Small molecules (HETATM)
62	- Metals (Zn, Fe, Mg, Ca, etc.)
63	- Cofactors (NAD, FAD, ATP)
64	- DNA/RNA
65
66	## Output format
67
68	```
69	output/
70	├── seqs/
71	│ └── protein.fa # FASTA sequences
72	└── protein_pdb/
73	└── protein_0001.pdb # PDBs with designed sequence
74	```
75
76	## Sample output
77
78	### Successful run
79	```
80	$ python run.py --pdb_path enzyme_substrate.pdb --out_folder output/ --num_seq_per_target 8
81	Loading LigandMPNN model weights...
82	Processing enzyme_substrate.pdb
83	Found ligand: LIG (12 atoms)
84	Generated 8 sequences in 3.1 seconds
85
86	output/seqs/enzyme_substrate.fa:
87	>enzyme_substrate_0001, score=1.45, global_score=1.38
88	MKTAYIAKQRQISFVKSHFSRQLE...
89	>enzyme_substrate_0002, score=1.52, global_score=1.41
90	MKTAYIAKQRQISFVKSQFSRQLD...
91	```
92
93	What good output looks like:
94	- Score: 1.0-2.0 (lower = more confident)
95	- Ligand detected and incorporated in context
96	- Active site residues preserved or optimized
97
98	## Decision tree
99
100	```
101	Should I use LigandMPNN?
102	│
103	├─ What's in your binding site?
104	│ ├─ Small molecule / ligand → LigandMPNN ✓
105	│ ├─ Metal ion (Zn, Fe, etc.) → LigandMPNN ✓
106	│ ├─ Cofactor (NAD, FAD, ATP) → LigandMPNN ✓
107	│ ├─ DNA/RNA → LigandMPNN ✓
108	│ └─ Nothing / protein only → Use ProteinMPNN
109	│
110	├─ What type of design?
111	│ ├─ Enzyme active site → LigandMPNN ✓
112	│ ├─ Metal binding site → LigandMPNN ✓
113	│ ├─ Protein-protein binder → Use ProteinMPNN
114	│ └─ De novo scaffold → Use ProteinMPNN
115	│
116	└─ Priority?
117	├─ Solubility/expression → Consider SolubleMPNN
118	└─ Ligand context accuracy → LigandMPNN ✓
119	```
120
121	## Typical performance
122
123	\| Campaign Size \| Time (T4) \| Cost (Modal) \| Notes \|
124	\|---------------\|-----------\|--------------\|-------\|
125	\| 100 backbones × 8 seq \| 15-20 min \| ~$2 \| Standard \|
126	\| 500 backbones × 8 seq \| 1-1.5h \| ~$8 \| Large campaign \|
127
128	Throughput: ~50-100 sequences/minute on T4 GPU.
129
130	---
131
132	## Verify
133
134	```bash
135	grep -c "^>" output/seqs/*.fa # Should match backbone_count × num_seq_per_target
136	```
137
138	---
139
140	## Troubleshooting
141
142	Ligand not recognized: Check HETATM format, verify ligand residue name
143	Poor binding residues: Increase sampling around active site
144	Missing contacts: Verify ligand coordinates in PDB
145
146	### Error interpretation
147
148	\| Error \| Cause \| Fix \|
149	\|-------\|-------\|-----\|
150	\| `RuntimeError: CUDA out of memory` \| Long protein or large batch \| Reduce batch_size \|
151	\| `KeyError: 'LIG'` \| Ligand not found in PDB \| Check HETATM records \|
152	\| `ValueError: no ligand atoms` \| Empty ligand \| Verify ligand has atoms in PDB \|
153
154	---
155
156	Next: Structure prediction for validation → `protein-qc` for filtering.

People who liked this also viewed

6 picks

Type

ligandmpnn

30installs20forks

View on GitHub

How to install ligandmpnn?

adaptyvbio/protein-design-skills/ligandmpnn

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

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# LigandMPNN Ligand-Aware 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	```bash
18	cd biomodals
19	# modal_ligandmpnn.py takes --input-pdb; LigandMPNN run.py args go in --params-str
20	modal run modal_ligandmpnn.py \
21	--input-pdb protein_ligand.pdb \
22	--params-str "--model_type ligand_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/LigandMPNN.git
30	cd LigandMPNN
31
32	python run.py \
33	--model_type ligand_mpnn \
34	--pdb_path protein_ligand.pdb \
35	--out_folder output/ \
36	--number_of_batches 16 \
37	--temperature 0.1
38	```
39
40	## Key parameters (LigandMPNN run.py)
41
42	\| Parameter \| Default \| Description \|
43	\|-----------\|---------\|-------------\|
44	\| `--pdb_path` \| required \| PDB with ligand \|
45	\| `--model_type` \| `protein_mpnn` \| `ligand_mpnn`, `soluble_mpnn`, etc. \|
46	\| `--temperature` \| 0.1 \| Sampling temperature \|
47	\| `--number_of_batches` \| 1 \| Batches (sequences = batch_size x batches) \|
48	\| `--batch_size` \| 1 \| Sequences per batch \|
49	\| `--ligand_mpnn_use_side_chain_context` \| 0 \| Use ligand side-chain context \|
50
51	## Ligand Specification
52
53	### In PDB File
54	Ligand must be present as HETATM records:
55	```
56	ATOM ...protein atoms...
57	HETATM 1 C1 LIG A 999 x.xxx y.yyy z.zzz 1.00 0.00 C
58	```
59
60	### Supported Ligand Types
61	- Small molecules (HETATM)
62	- Metals (Zn, Fe, Mg, Ca, etc.)
63	- Cofactors (NAD, FAD, ATP)
64	- DNA/RNA
65
66	## Output format
67
68	```
69	output/
70	├── seqs/
71	│ └── protein.fa # FASTA sequences
72	└── protein_pdb/
73	└── protein_0001.pdb # PDBs with designed sequence
74	```
75
76	## Sample output
77
78	### Successful run
79	```
80	$ python run.py --pdb_path enzyme_substrate.pdb --out_folder output/ --num_seq_per_target 8
81	Loading LigandMPNN model weights...
82	Processing enzyme_substrate.pdb
83	Found ligand: LIG (12 atoms)
84	Generated 8 sequences in 3.1 seconds
85
86	output/seqs/enzyme_substrate.fa:
87	>enzyme_substrate_0001, score=1.45, global_score=1.38
88	MKTAYIAKQRQISFVKSHFSRQLE...
89	>enzyme_substrate_0002, score=1.52, global_score=1.41
90	MKTAYIAKQRQISFVKSQFSRQLD...
91	```
92
93	What good output looks like:
94	- Score: 1.0-2.0 (lower = more confident)
95	- Ligand detected and incorporated in context
96	- Active site residues preserved or optimized
97
98	## Decision tree
99
100	```
101	Should I use LigandMPNN?
102	│
103	├─ What's in your binding site?
104	│ ├─ Small molecule / ligand → LigandMPNN ✓
105	│ ├─ Metal ion (Zn, Fe, etc.) → LigandMPNN ✓
106	│ ├─ Cofactor (NAD, FAD, ATP) → LigandMPNN ✓
107	│ ├─ DNA/RNA → LigandMPNN ✓
108	│ └─ Nothing / protein only → Use ProteinMPNN
109	│
110	├─ What type of design?
111	│ ├─ Enzyme active site → LigandMPNN ✓
112	│ ├─ Metal binding site → LigandMPNN ✓
113	│ ├─ Protein-protein binder → Use ProteinMPNN
114	│ └─ De novo scaffold → Use ProteinMPNN
115	│
116	└─ Priority?
117	├─ Solubility/expression → Consider SolubleMPNN
118	└─ Ligand context accuracy → LigandMPNN ✓
119	```
120
121	## Typical performance
122
123	\| Campaign Size \| Time (T4) \| Cost (Modal) \| Notes \|
124	\|---------------\|-----------\|--------------\|-------\|
125	\| 100 backbones × 8 seq \| 15-20 min \| ~$2 \| Standard \|
126	\| 500 backbones × 8 seq \| 1-1.5h \| ~$8 \| Large campaign \|
127
128	Throughput: ~50-100 sequences/minute on T4 GPU.
129
130	---
131
132	## Verify
133
134	```bash
135	grep -c "^>" output/seqs/*.fa # Should match backbone_count × num_seq_per_target
136	```
137
138	---
139
140	## Troubleshooting
141
142	Ligand not recognized: Check HETATM format, verify ligand residue name
143	Poor binding residues: Increase sampling around active site
144	Missing contacts: Verify ligand coordinates in PDB
145
146	### Error interpretation
147
148	\| Error \| Cause \| Fix \|
149	\|-------\|-------\|-----\|
150	\| `RuntimeError: CUDA out of memory` \| Long protein or large batch \| Reduce batch_size \|
151	\| `KeyError: 'LIG'` \| Ligand not found in PDB \| Check HETATM records \|
152	\| `ValueError: no ligand atoms` \| Empty ligand \| Verify ligand has atoms in PDB \|
153
154	---
155
156	Next: Structure prediction for validation → `protein-qc` for filtering.

People who liked this also viewed

6 picks

Type