germinal

Name: germinal
Author: adaptyvbio

7installs20forks

How to install germinal?

adaptyvbio/protein-design-skills/germinal

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

Installs into the current project.

Summary

De novo antibody and nanobody (VHH) design with Germinal. Use this skill when: (1) Designing epitope-targeted nanobodies or scFvs, (2) Needing CDR design on a fixed framework, (3) Working on antibody-format binders rather than miniproteins. For miniprotein binders, use binder-des

Files · 1

View on GitHub

SKILL.md

1	# Germinal Antibody and Nanobody Design
2
3	[Germinal](https://github.com/SantiagoMille/germinal) is an open pipeline for
4	epitope-targeted de novo antibody and nanobody design. It hallucinates CDRs on a
5	fixed framework, designs sequences with AbMPNN, and cofolds with a structure
6	predictor (it downloads AlphaFold-Multimer params). Runnable through biomodals.
7
8	The biomodals author notes Germinal is finicky and suggests BoltzGen for general
9	binder design; treat Germinal as the antibody-format option, not a default.
10
11	## Prerequisites
12
13	\| Requirement \| Value \|
14	\|-------------\|-------\|
15	\| Runner \| Modal (biomodals) \|
16	\| GPU \| H100 (default; `GPU` env var) \|
17	\| Setup \| See [Getting started](../../docs/getting-started.md) \|
18
19	## How to run
20
21	```bash
22	git clone https://github.com/hgbrian/biomodals && cd biomodals
23
24	uv run --with modal --with PyYAML modal run modal_germinal.py \
25	--target-yaml target_example.yaml \
26	--max-trajectories 1 \
27	--max-passing-designs 1
28	```
29
30	## Key parameters
31
32	\| Parameter \| Default \| Description \|
33	\|-----------\|---------\|-------------\|
34	\| `--target-yaml` \| required \| Target config (target_name, target_pdb_path, target_chain, binder_chain, target_hotspots, length) \|
35	\| `--run-type` \| `vhh` \| `vhh` (nanobody) or `scfv` \|
36	\| `--max-trajectories` \| 100 \| Trajectories to run \|
37	\| `--max-passing-designs` \| 10 \| Stop after this many passing designs \|
38	\| `--out-dir` \| `./out/germinal` \| Output directory \|
39
40	## Target YAML
41
42	```yaml
43	target_name: PDL1
44	target_pdb_path: target.pdb
45	target_chain: A
46	binder_chain: B
47	target_hotspots: "45,67,89"
48	length: 120
49	```
50
51	## Decision tree
52
53	```
54	Antibody-format binder?
55	│
56	├─ Nanobody / VHH → germinal (run-type vhh) or mber
57	├─ scFv → germinal (run-type scfv)
58	└─ Miniprotein (not antibody) → binder-design (boltzgen, bindcraft, mosaic)
59	```
60
61	For VHH nanobodies, biomodals also has `modal_mber.py` (mBER) and `modal_iggm.py`
62	(IgGM) as alternatives.
63
64	## Cost
65
66	Adaptyv's own tests of these models showed Germinal costing about $1.60 per accepted
67	design, averaged across 7 targets.
68
69	## Troubleshooting
70
71	\| Issue \| Cause \| Fix \|
72	\|-------\|-------\|-----\|
73	\| Pipeline fails early \| Missing PyYAML \| Add `--with PyYAML` to the invocation \|
74	\| No passing designs \| Hard epitope or low budget \| Raise `--max-trajectories` \|
75	\| OOM \| Large target \| Use the default H100 or trim the target \|
76
77	---
78
79	Next: Validate with `boltz` or `chai`, rank with `ipsae`, filter with `protein-qc`.

People who liked this also viewed

6 picks

Type

germinal

7installs20forks

View on GitHub

How to install germinal?

adaptyvbio/protein-design-skills/germinal

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

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Germinal Antibody and Nanobody Design
2
3	[Germinal](https://github.com/SantiagoMille/germinal) is an open pipeline for
4	epitope-targeted de novo antibody and nanobody design. It hallucinates CDRs on a
5	fixed framework, designs sequences with AbMPNN, and cofolds with a structure
6	predictor (it downloads AlphaFold-Multimer params). Runnable through biomodals.
7
8	The biomodals author notes Germinal is finicky and suggests BoltzGen for general
9	binder design; treat Germinal as the antibody-format option, not a default.
10
11	## Prerequisites
12
13	\| Requirement \| Value \|
14	\|-------------\|-------\|
15	\| Runner \| Modal (biomodals) \|
16	\| GPU \| H100 (default; `GPU` env var) \|
17	\| Setup \| See [Getting started](../../docs/getting-started.md) \|
18
19	## How to run
20
21	```bash
22	git clone https://github.com/hgbrian/biomodals && cd biomodals
23
24	uv run --with modal --with PyYAML modal run modal_germinal.py \
25	--target-yaml target_example.yaml \
26	--max-trajectories 1 \
27	--max-passing-designs 1
28	```
29
30	## Key parameters
31
32	\| Parameter \| Default \| Description \|
33	\|-----------\|---------\|-------------\|
34	\| `--target-yaml` \| required \| Target config (target_name, target_pdb_path, target_chain, binder_chain, target_hotspots, length) \|
35	\| `--run-type` \| `vhh` \| `vhh` (nanobody) or `scfv` \|
36	\| `--max-trajectories` \| 100 \| Trajectories to run \|
37	\| `--max-passing-designs` \| 10 \| Stop after this many passing designs \|
38	\| `--out-dir` \| `./out/germinal` \| Output directory \|
39
40	## Target YAML
41
42	```yaml
43	target_name: PDL1
44	target_pdb_path: target.pdb
45	target_chain: A
46	binder_chain: B
47	target_hotspots: "45,67,89"
48	length: 120
49	```
50
51	## Decision tree
52
53	```
54	Antibody-format binder?
55	│
56	├─ Nanobody / VHH → germinal (run-type vhh) or mber
57	├─ scFv → germinal (run-type scfv)
58	└─ Miniprotein (not antibody) → binder-design (boltzgen, bindcraft, mosaic)
59	```
60
61	For VHH nanobodies, biomodals also has `modal_mber.py` (mBER) and `modal_iggm.py`
62	(IgGM) as alternatives.
63
64	## Cost
65
66	Adaptyv's own tests of these models showed Germinal costing about $1.60 per accepted
67	design, averaged across 7 targets.
68
69	## Troubleshooting
70
71	\| Issue \| Cause \| Fix \|
72	\|-------\|-------\|-----\|
73	\| Pipeline fails early \| Missing PyYAML \| Add `--with PyYAML` to the invocation \|
74	\| No passing designs \| Hard epitope or low budget \| Raise `--max-trajectories` \|
75	\| OOM \| Large target \| Use the default H100 or trim the target \|
76
77	---
78
79	Next: Validate with `boltz` or `chai`, rank with `ipsae`, filter with `protein-qc`.

People who liked this also viewed

6 picks

Type