binder-design

Name: binder-design
Author: adaptyvbio

41installs20forks

How to install binder-design?

adaptyvbio/protein-design-skills/binder-design

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

Installs into the current project.

Summary

Guidance for choosing the right protein binder design tool. Use this skill when: (1) Deciding between BoltzGen, BindCraft, or RFdiffusion, (2) Planning a binder design campaign, (3) Understanding trade-offs between different approaches, (4) Selecting tools for specific target typ

Files · 1

View on GitHub

SKILL.md

1	# Binder Design Tool Selection
2
3	## Which tool wins
4
5	No single tool is best for every target. Hit-rate is strongly target-dependent, so
6	choose by target type, what you want to control, and available compute.
7
8	The clearest signal comes from head-to-head competitions where many methods design
9	against the same target. On the Adaptyv Nipah de novo target, the public results show:
10
11	\| Method \| Tested \| Binders \| Hit-rate \|
12	\|--------\|--------\|---------\|----------\|
13	\| Mosaic (gradient, multi-model) \| 9 \| 8 \| 89% \|
14	\| ProteinMPNN hybrid \| 28 \| 7 \| 25% \|
15	\| RFdiffusion \| 60 \| 13 \| 22% \|
16	\| BindCraft \| 98 \| 7 \| 7% \|
17	\| BoltzGen \| 182 \| 6 \| 3% \|
18
19	Mosaic had the highest hit-rate here, but on a small, expert-tuned sample. The ranking
20	shifts on other targets, and that target-dependence is true of every method (BoltzGen,
21	Boltz, BindCraft, Mosaic). You cannot know a priori which will win on a new target, so
22	this is not a fixed leaderboard.
23
24	Because of that, choose a starting point by cost and effort to a binder, not by
25	assuming a method has the best hit-rate. BoltzGen is the suggested default because it is
26	turnkey and all-atom, so it gets you testable designs fastest with the least setup.
27	Mosaic is the high-ceiling option when you can invest time tuning the objective. On a
28	hard or important target, running more than one method in parallel is reasonable.
29
30	```
31	De novo binder design?
32	│
33	├─ Lowest cost/effort to testable designs → BoltzGen (default)
34	├─ Hard/important target, can invest tuning → Mosaic (gradient, multi-model)
35	├─ Ligand / small-molecule binding → BoltzGen (all-atom)
36	├─ Diversity / exploration → RFdiffusion + ProteinMPNN
37	├─ End-to-end with built-in validation → BindCraft
38	└─ Antibody / nanobody (VHH) → germinal skill (also mber, iggm in biomodals)
39	```
40
41	## Tool comparison
42
43	\| Tool \| Strengths \| Weaknesses \| Best for \|
44	\|------\|-----------\|------------\|----------\|
45	\| BoltzGen \| All-atom, single-step, turnkey \| One model in the loop; mid-range cost per design \| Lowest-effort default, ligand binding \|
46	\| Mosaic \| Composable multi-model objective, won hard head-to-heads \| Needs tuning, local JAX only \| Hard or important targets, expert use \|
47	\| BindCraft \| End-to-end, built-in AF2 validation \| Less diverse \| Production campaigns \|
48	\| RFdiffusion \| High diversity \| Requires ProteinMPNN; not in biomodals \| Exploration, diversity \|
49	\| Germinal \| Antibody and nanobody formats \| Finicky \| scFv / VHH design \|
50
51	## Compute cost per design
52
53	Adaptyv's own tests of these models showed the following compute cost per accepted
54	design, averaged across 7 targets (it varies several-fold by target):
55
56	\| Method \| Cost per design \|
57	\|--------\|-----------------\|
58	\| RSO \| ~$0.15 \|
59	\| RFdiffusion \| ~$0.25 \|
60	\| Mosaic \| ~$0.55 \|
61	\| ESMFold2 inversion \| ~$0.85 \|
62	\| mBER \| ~$1.40 \|
63	\| Germinal \| ~$1.60 \|
64	\| BoltzGen \| ~$1.80 \|
65	\| BindCraft \| ~$2.90 \|
66
67	Per-design compute cost is not the same as cost to a binder, which also depends on the
68	hit-rate on your target. The gradient methods (RSO, Mosaic) are cheap per design but
69	need setup and tuning; BoltzGen and BindCraft cost more per design but are turnkey, so
70	their advantage is low human effort rather than lowest compute cost.
71
72	## Compute vs effort tradeoff
73
74	- Lowest human effort: BoltzGen needs no tuning and runs through biomodals. Good
75	first pass and good for ligand binding.
76	- Highest ceiling on a hard target: Mosaic, given time to design and tune the
77	objective. It runs locally on a JAX GPU rather than through biomodals, and is cheap
78	per design.
79	- Whatever the generator, validate with `boltz` or `chai` and rank with `ipsae`.
80
81	Other biomodals-backed options: `modal_rso.py` (Rejection Sampling Optimization, an
82	AlphaFold-based gradient method) for minibinders, and `modal_mber.py` for VHH
83	nanobodies.
84
85	## Example pipeline: BoltzGen → Chai → QC
86
87	BoltzGen provides all-atom design with built-in side-chain packing. This is one
88	turnkey path; swap in Mosaic, RFdiffusion, or BindCraft depending on the target.
89
90	```
91	Target → BoltzGen → Validate → Filter
92	(pdb) (all-atom) (chai) (qc)
93	```
94
95	### 1. Target preparation
96	```bash
97	# Fetch structure from PDB
98	# Use pdb skill for guidance
99	```
100	- Trim to binding region + 10A buffer
101	- Remove waters and ligands
102	- Renumber chains if needed
103
104	### 2. Hotspot selection
105	- Choose 3-6 exposed residues
106	- Prefer charged/aromatic residues
107	- Cluster spatially (within 10-15A)
108
109	### 3. Design with BoltzGen
110
111	First, create a YAML config file (e.g., `binder.yaml`):
112	```yaml
113	entit

People who liked this also viewed

6 picks

Type

binder-design

41installs20forks

View on GitHub

How to install binder-design?

adaptyvbio/protein-design-skills/binder-design

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

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Binder Design Tool Selection
2
3	## Which tool wins
4
5	No single tool is best for every target. Hit-rate is strongly target-dependent, so
6	choose by target type, what you want to control, and available compute.
7
8	The clearest signal comes from head-to-head competitions where many methods design
9	against the same target. On the Adaptyv Nipah de novo target, the public results show:
10
11	\| Method \| Tested \| Binders \| Hit-rate \|
12	\|--------\|--------\|---------\|----------\|
13	\| Mosaic (gradient, multi-model) \| 9 \| 8 \| 89% \|
14	\| ProteinMPNN hybrid \| 28 \| 7 \| 25% \|
15	\| RFdiffusion \| 60 \| 13 \| 22% \|
16	\| BindCraft \| 98 \| 7 \| 7% \|
17	\| BoltzGen \| 182 \| 6 \| 3% \|
18
19	Mosaic had the highest hit-rate here, but on a small, expert-tuned sample. The ranking
20	shifts on other targets, and that target-dependence is true of every method (BoltzGen,
21	Boltz, BindCraft, Mosaic). You cannot know a priori which will win on a new target, so
22	this is not a fixed leaderboard.
23
24	Because of that, choose a starting point by cost and effort to a binder, not by
25	assuming a method has the best hit-rate. BoltzGen is the suggested default because it is
26	turnkey and all-atom, so it gets you testable designs fastest with the least setup.
27	Mosaic is the high-ceiling option when you can invest time tuning the objective. On a
28	hard or important target, running more than one method in parallel is reasonable.
29
30	```
31	De novo binder design?
32	│
33	├─ Lowest cost/effort to testable designs → BoltzGen (default)
34	├─ Hard/important target, can invest tuning → Mosaic (gradient, multi-model)
35	├─ Ligand / small-molecule binding → BoltzGen (all-atom)
36	├─ Diversity / exploration → RFdiffusion + ProteinMPNN
37	├─ End-to-end with built-in validation → BindCraft
38	└─ Antibody / nanobody (VHH) → germinal skill (also mber, iggm in biomodals)
39	```
40
41	## Tool comparison
42
43	\| Tool \| Strengths \| Weaknesses \| Best for \|
44	\|------\|-----------\|------------\|----------\|
45	\| BoltzGen \| All-atom, single-step, turnkey \| One model in the loop; mid-range cost per design \| Lowest-effort default, ligand binding \|
46	\| Mosaic \| Composable multi-model objective, won hard head-to-heads \| Needs tuning, local JAX only \| Hard or important targets, expert use \|
47	\| BindCraft \| End-to-end, built-in AF2 validation \| Less diverse \| Production campaigns \|
48	\| RFdiffusion \| High diversity \| Requires ProteinMPNN; not in biomodals \| Exploration, diversity \|
49	\| Germinal \| Antibody and nanobody formats \| Finicky \| scFv / VHH design \|
50
51	## Compute cost per design
52
53	Adaptyv's own tests of these models showed the following compute cost per accepted
54	design, averaged across 7 targets (it varies several-fold by target):
55
56	\| Method \| Cost per design \|
57	\|--------\|-----------------\|
58	\| RSO \| ~$0.15 \|
59	\| RFdiffusion \| ~$0.25 \|
60	\| Mosaic \| ~$0.55 \|
61	\| ESMFold2 inversion \| ~$0.85 \|
62	\| mBER \| ~$1.40 \|
63	\| Germinal \| ~$1.60 \|
64	\| BoltzGen \| ~$1.80 \|
65	\| BindCraft \| ~$2.90 \|
66
67	Per-design compute cost is not the same as cost to a binder, which also depends on the
68	hit-rate on your target. The gradient methods (RSO, Mosaic) are cheap per design but
69	need setup and tuning; BoltzGen and BindCraft cost more per design but are turnkey, so
70	their advantage is low human effort rather than lowest compute cost.
71
72	## Compute vs effort tradeoff
73
74	- Lowest human effort: BoltzGen needs no tuning and runs through biomodals. Good
75	first pass and good for ligand binding.
76	- Highest ceiling on a hard target: Mosaic, given time to design and tune the
77	objective. It runs locally on a JAX GPU rather than through biomodals, and is cheap
78	per design.
79	- Whatever the generator, validate with `boltz` or `chai` and rank with `ipsae`.
80
81	Other biomodals-backed options: `modal_rso.py` (Rejection Sampling Optimization, an
82	AlphaFold-based gradient method) for minibinders, and `modal_mber.py` for VHH
83	nanobodies.
84
85	## Example pipeline: BoltzGen → Chai → QC
86
87	BoltzGen provides all-atom design with built-in side-chain packing. This is one
88	turnkey path; swap in Mosaic, RFdiffusion, or BindCraft depending on the target.
89
90	```
91	Target → BoltzGen → Validate → Filter
92	(pdb) (all-atom) (chai) (qc)
93	```
94
95	### 1. Target preparation
96	```bash
97	# Fetch structure from PDB
98	# Use pdb skill for guidance
99	```
100	- Trim to binding region + 10A buffer
101	- Remove waters and ligands
102	- Renumber chains if needed
103
104	### 2. Hotspot selection
105	- Choose 3-6 exposed residues
106	- Prefer charged/aromatic residues
107	- Cluster spatially (within 10-15A)
108
109	### 3. Design with BoltzGen
110
111	First, create a YAML config file (e.g., `binder.yaml`):
112	```yaml
113	entit

People who liked this also viewed

6 picks

Type