binder-design
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 types. For specific tool parameters, use the individual tool skills (boltzgen, bindcraft, rfdiffusion, etc.).
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 6
- Forks
- 0
01 前言 Front Matter
- name
- binder-design
- tags
- [ "guidance", "tool-selection", "workflow" ]
- license
- MIT
- category
- orchestration
- description
- 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 types. For specific tool parameters, use the individual tool skills (boltzgen, bindcraft, rfdiffusion, etc.).
03 SKILL.md 正文
Binder Design Tool Selection
Which tool wins
No single tool is best for every target. Hit-rate is strongly target-dependent, so choose by target type, what you want to control, and available compute.
The clearest signal comes from head-to-head competitions where many methods design against the same target. On the Adaptyv Nipah de novo target, the public results show:
| Method | Tested | Binders | Hit-rate |
|---|---|---|---|
| Mosaic (gradient, multi-model) | 9 | 8 | 89% |
| ProteinMPNN hybrid | 28 | 7 | 25% |
| RFdiffusion | 60 | 13 | 22% |
| BindCraft | 98 | 7 | 7% |
| BoltzGen | 182 | 6 | 3% |
Mosaic had the highest hit-rate here, but on a small, expert-tuned sample. The ranking shifts on other targets, and that target-dependence is true of every method (BoltzGen, Boltz, BindCraft, Mosaic). You cannot know a priori which will win on a new target, so this is not a fixed leaderboard.
Because of that, choose a starting point by cost and effort to a binder, not by assuming a method has the best hit-rate. BoltzGen is the suggested default because it is turnkey and all-atom, so it gets you testable designs fastest with the least setup. Mosaic is the high-ceiling option when you can invest time tuning the objective. On a hard or important target, running more than one method in parallel is reasonable.
De novo binder design?
│
├─ Lowest cost/effort to testable designs → BoltzGen (default)
├─ Hard/important target, can invest tuning → Mosaic (gradient, multi-model)
├─ Ligand / small-molecule binding → BoltzGen (all-atom)
├─ Diversity / exploration → RFdiffusion + ProteinMPNN
├─ End-to-end with built-in validation → BindCraft
└─ Antibody / nanobody (VHH) → germinal skill (also mber, iggm in biomodals)
Tool comparison
| Tool | Strengths | Weaknesses | Best for |
|---|---|---|---|
| BoltzGen | All-atom, single-step, turnkey | One model in the loop; mid-range cost per design | Lowest-effort default, ligand binding |
| Mosaic | Composable multi-model objective, won hard head-to-heads | Needs tuning, local JAX only | Hard or important targets, expert use |
| BindCraft | End-to-end, built-in AF2 validation | Less diverse | Production campaigns |
| RFdiffusion | High diversity | Requires ProteinMPNN; not in biomodals | Exploration, diversity |
| Germinal | Antibody and nanobody formats | Finicky | scFv / VHH design |
Compute cost per design
Adaptyv's own tests of these models showed the following compute cost per accepted design, averaged across 7 targets (it varies several-fold by target):
| Method | Cost per design |
|---|---|
| RSO | ~$0.15 |
| RFdiffusion | ~$0.25 |
| Mosaic | ~$0.55 |
| ESMFold2 inversion | ~$0.85 |
| mBER | ~$1.40 |
| Germinal | ~$1.60 |
| BoltzGen | ~$1.80 |
| BindCraft | ~$2.90 |
Per-design compute cost is not the same as cost to a binder, which also depends on the hit-rate on your target. The gradient methods (RSO, Mosaic) are cheap per design but need setup and tuning; BoltzGen and BindCraft cost more per design but are turnkey, so their advantage is low human effort rather than lowest compute cost.
Compute vs effort tradeoff
- Lowest human effort: BoltzGen needs no tuning and runs through biomodals. Good first pass and good for ligand binding.
- Highest ceiling on a hard target: Mosaic, given time to design and tune the objective. It runs locally on a JAX GPU rather than through biomodals, and is cheap per design.
- Whatever the generator, validate with
boltzorchaiand rank withipsae.
Other biomodals-backed options: modal_rso.py (Rejection Sampling Optimization, an
AlphaFold-based gradient method) for minibinders, and modal_mber.py for VHH
nanobodies.
Example pipeline: BoltzGen → Chai → QC
BoltzGen provides all-atom design with built-in side-chain packing. This is one turnkey path; swap in Mosaic, RFdiffusion, or BindCraft depending on the target.
Target → BoltzGen → Validate → Filter
(pdb) (all-atom) (chai) (qc)
1. Target preparation
# Fetch structure from PDB
# Use pdb skill for guidance
- Trim to binding region + 10A buffer
- Remove waters and ligands
- Renumber chains if needed
2. Hotspot selection
- Choose 3-6 exposed residues
- Prefer charged/aromatic residues
- Cluster spatially (within 10-15A)
3. Design with BoltzGen
First, create a YAML config file (e.g., binder.yaml):
entities:
- protein:
id: B
sequence: 70..100
- file:
path: target.cif
include:
- chain:
id: A
binding_types:
- chain:
id: A
binding: 45,67,89
Then run:
modal run modal_boltzgen.py \
--input-yaml binder.yaml \
--protocol protein-anything \
--num-designs 50
Why BoltzGen?
- All-atom output (no separate ProteinMPNN step needed)
- Better for ligand/small molecule binding
- Single-step design (backbone + sequence + side chains)
4. Alternative: RFdiffusion Pipeline
For maximum diversity or when backbone-only is preferred:
# Step 1: Backbone generation (RFdiffusion, run from the official repo)
python run_inference.py \
inference.input_pdb=target.pdb \
contigmap.contigs=[A1-150/0 70-100] \
ppi.hotspot_res=[A45,A67,A89] \
inference.num_designs=500
# Step 2: Sequence design
modal run modal_ligandmpnn.py \
--input-pdb backbone.pdb \
--params-str "--number_of_batches 16 --temperature 0.1"
5. Validation
modal run modal_chai1.py \
--input-faa sequences.fasta \
--out-dir predictions/
6. Filtering
Apply standard thresholds:
- pLDDT > 0.80
- ipTM > 0.50
- PAE_interface < 10
- scRMSD < 2.0 A
See protein-qc skill for details.
Number of designs
| Stage | Count | Purpose |
|---|---|---|
| Backbone generation | 500-1000 | Diversity |
| Sequences per backbone | 8-16 | Sequence space |
| AF2 predictions | All | Validation |
| After filtering | 50-200 | Candidates |
| Experimental testing | 10-50 | Final selection |
Common mistakes
Wrong hotspots
- Using buried residues
- Too many hotspots (over-constrain)
- Wrong chain/residue numbers
Insufficient diversity
- Too few designs generated
- Low temperature in ProteinMPNN
- Not exploring multiple backbones
Poor target preparation
- Including full protein instead of binding region
- Missing important structural features
- Wrong protonation states
Timeline guide
| Step | Compute Time |
|---|---|
| RFdiffusion (500 designs) | 2-4 hours |
| ProteinMPNN (8000 sequences) | 1-2 hours |
| AF2 prediction (8000 sequences) | 12-24 hours |
| Filtering and analysis | 1-2 hours |
Total: 1-2 days of compute
04 同仓库更多 Skills
serpmaven
Scrape Google search results (SERP) to JSON with SERP Maven (serpmaven) - a real Chrome browser on a residential proxy with sticky sessions. Run it as a CLI, a Python library, a local HTTP search API (`serve`), an MCP server or a Docker image; set the proxy login, workers, pages and cost tracking. Use when the user wants Google results, rank tracking, a self-hosted SERP API, or live Google search for an AI agent.
revnu-landing-page-audit
Find out why a landing page isn't turning visitors into signups, demos or sales. Audits one page against what the ad, post or search that sends people there promised, then checks the first screen, proof, call to action, friction and mobile. Returns at most 7 prioritized fixes, plus what is fine and what could not be checked; rewrites specific lines and fixes the page in the repo when asked. Never promises a conversion lift. Works without a Revnu account. Use when the user says or asks: "audit my landing page", "why isn't this page converting", "review my homepage", "my ads get clicks but no signups", "people bounce from the pricing page", "CRO review", "improve conversion rate", "is my hero section clear", "roast my landing page", "rewrite my headline", "what's wrong with this page", "landing page feedback", "make the signup page convert better". Not for SEO or AI-search audits, page speed engineering on its own, or building a new page from scratch.
revnu-growth-context
Write down once who buys, what they're stuck on, what it costs, what proof you have and what you've already tried, so every growth task starts from facts instead of a questionnaire. Creates or updates .agents/revnu-growth.md from the repo, the live site and the user, and imports an existing product-marketing.md. Use when the user says "set up our marketing context", "who is our ideal customer", "write our ICP", "describe the product for marketing", "positioning doc", "what's our brand voice", "remember this about our customers", "log what we tried", "what have we learned from marketing", "we changed pricing, update the context", "update our growth notes", "summarize the business for the agent", "import my product-marketing.md", "start a growth doc", or before a first marketing, SEO, outreach or launch task when no context file exists. Works without a Revnu account.
revnu-get-listed
Get onto the third-party pages AI answers and search cite when buyers ask for the best tool in your category: "best X" roundups, directories, review sites, GitHub awesome lists, community threads and comparison pages. Finds which ones are cited, ranks them by fit and effort, and drafts each submission, PR or pitch by the target's own rules. Use when the user says "get us listed", "which directories should we submit to", "how do we get into best-of lists", "why does ChatGPT recommend competitors and not us", "get mentioned in AI answers", "submit to awesome lists", "get on G2", "where are our competitors listed", "pitch listicle authors", "find backlink opportunities", "launch directories", "get into roundups", "Product Hunt alternatives", "where should we be listed", "get cited by Perplexity". Never submits, pays or posts without a yes. Works without a Revnu account.
revnu-ai-readiness-audit
Audit whether your website gives ChatGPT, Perplexity, Claude, Gemini and Google's AI Overviews something worth quoting, and fix the pages that don't. Use when the user says or asks "is my site ready for AI search", "audit my site for AI answers", "AEO audit", "GEO audit", "why doesn't ChatGPT quote our pricing", "make our pages citable by AI", "do I need llms.txt", "does schema help with ChatGPT", "check our structured data", "is our content visible without JavaScript", "AI SEO audit", "optimize our homepage for AI answers", "LLM SEO check", "why does AI get our product wrong", "fix our site for AI Overviews". Checks fetchability and indexability, answers in the raw HTML, entity facts, schema (including JS-injected markup), headings, dates and llms.txt, then gives at most 10 prioritised fixes and edits the repo when asked. Works without a Revnu account.
proxy-troubleshooting
Diagnose residential proxy failures, with NodeMaven's gateway as the reference - 407 with correct credentials, 406, 410, 500, a connection that hangs, ERR_TUNNEL_CONNECTION_FAILED in a browser, Playwright or Puppeteer ignoring proxy credentials, a sticky session that changes IP, geo-targeting that is silently ignored, passwords with special characters, and unexpected traffic from headless Chrome. Use when a proxy connection fails, returns the wrong location, or costs more traffic than expected.