// Agentic AI · Protein Design · Drug Discovery
AminoClaw
The AI Co-Scientist Every Biologist Deserves
Brilliant biological ideas meet frontier AI execution. AminoClaw automates the entire protein design pipeline — from literature to optimized binders — so your hypotheses don't wait months to become results.
0.00
ipSAE Score
RBX1 binder (0.30 → 0.65)
10⁵×
Selectivity Gain
SlUGT enzyme redesign
0
Bio Skills
LabClaw ecosystem
RBX1 Case Study → See Results ↓
SCROLL
// What We've Already Built
Real Results, Real Science

Three projects. Three domains. All automated by AminoClaw's agentic pipeline.

🏆 ProteinBASE Competition
0.30
RBX1 Binder Design
Targeting the E2-binding surface of RBX1 RING domain — an E3 ubiquitin ligase implicated in cancer. Agent read 50+ papers, identified key interface residues, then drove gradient optimization through AlphaFold3.
Before (random init)After (AF3 + Mosaic)
ipSAE 0.30ipSAE 0.65
🧪 DMS Analysis · LCC
25
LCC Enzyme Optimization
Leaf-Branch Compost Cutinase for industrial PET recycling. Agent analyzed 1,247 single mutants and 8,179 multi-site variants, mapping the complete fitness landscape and identifying elite hotspot positions.
+1.84×
Top mutation N215H
9,426
Variants analyzed
3
Coldspot positions
⚗️ Enzyme Engineering
10⁵×
SlUGT91R1 Specificity Switch
Switching UDP-arabinose:rosin arabinosyltransferase from UDP-Arap to UDP-Araf specificity. Agent designed and ranked 6 mutants using RF3 structure prediction + AutoDock Vina, identifying M6 as the optimal variant.
M6 ★ TOP CANDIDATE
Y386A / Q389A / I20F / G290S
ΔSel vs WT: +9.944 kcal/mol
See Pipeline →
// The Gap We're Closing
Brilliant Ideas, Blocked by Tools
Biologists spend years mastering biology. They shouldn't need to master AlphaFold3, ProteinMPNN, AutoDock Vina, PDB APIs, and Python scripting just to test an idea.
🧬
The Tools Exist — But Aren't Accessible
AlphaFold3, Boltz-2, ProteinMPNN, RF3 are state-of-the-art. But each requires specialized setup, coding, and interpretation. Most biologists can't access them without a computational collaborator.
⚙️
Fragmented, Manual Pipeline
Literature → PDB → structure prediction → docking → analysis → report: each step is a separate tool, a separate skillset, a separate week. No one has assembled this into a coherent workflow.
⏳
From Idea to Result Takes Months
A single design cycle — literature review, target analysis, structure prediction, binder design, scoring — typically takes 3–6 months with a computational team. AminoClaw does it in hours.
Traditional vs. AminoClaw
Literature review
2–4 weeks
Structural data collection
1–2 weeks
Hotspot / interface analysis
2–4 weeks
Computational design runs
4–8 weeks
Analysis & reporting
2–3 weeks
Total: 3–5 months ❌
Agent reads literature & databases
~30 min
Hotspot identification + DMS analysis
~1 hour
AI model invocation + design
2–4 hours
Interactive report generated
~5 min
Total: <1 day ✓
// How AminoClaw Works
End-to-End Agentic Pipeline

One instruction. Six automated stages. Hover each step to see the tools invoked.

📚
01
Literature Scan
PubMed · Semantic Scholar · ChemRxiv · bioRxiv
🗄️
02
Data Retrieval
PDB · UniProt · AlphaFold DB · ProteinBASE · BindingDB
📊
03
Data Analysis
DMS parsing · fitness scoring · hotspot detection · epistasis mapping
🎯
04
Residue Mapping
Interface prediction · conservation analysis · functional site annotation
⚙️
05
Model Invocation
AlphaFold3 · ProteinMPNN · Boltz-2 · AutoDock Vina · RoseTTAFold3
📋
06
Report Generation
Interactive HTML · 3D structure viewer · figures · actionable recommendations
// Case Study 01 · RBX1 Binder Design
E3 Ligase Inhibition via AI-Designed Binder
RBX1 RING domain recruits E2 ubiquitin-conjugating enzymes. Blocking this interface could suppress oncogenic CRL-mediated ubiquitination of tumor suppressors (p27, CDT1).
1
Agent reads 50+ papers on RBX1, CRL complexes, RING domain biology Identifies E2-binding interface: L2 loop (I53–A57) and RING loop (Q65–I69)
2
Retrieves 2LGV crystal structure from PDB RBX1-UBE2D2 complex — provides interface geometry for binder design
3
Runs Proteina-complexa + Germinal via autoresearch pipeline Automated multi-round design · candidate generation · structure scoring
4
Rescores top candidates with Boltz-2 & Protenix Multi-model consensus scoring for robust ranking
5
Generates interactive report with 3D structure viewer Top 10 binders ranked by ipTM, pLDDT, PAE — publication-ready
Binder Quality Improvement
0.30
ipSAE Initial
→
0.65
ipSAE Optimized
+116%
ipSAE improvement via AF3 gradient optimization. Final ipTM: 0.83. Binder confidently predicted to engage the E2-interface of RBX1.
RBX1 RING DOMAIN — interface residues highlighted
MGSSHHHHHH SSGLVPRGSH MGSHMAQPEL DTVDMKGCDVLG QCIFHKWLQGMCAACVNQLNFVNAAKREITELT
■L2 loop (E2 contact)
■RING loop (E2 recruit)
Explore full binder report → rbx1.acetyl.online
// LCC Mutational Fitness Landscape
1,247 single mutants · 43 positions shown · 6 aa substitutions per column
Hotspot (beneficial)
Coldspot (harmful)
Neutral
Pos 1Pos 83 ★Pos 183Pos 258
// Case Study 02 · LCC Enzyme Optimization
From Raw DMS Data to Engineering Roadmap
Agent analyzed the complete DMS landscape of Leaf-Branch Compost Cutinase — a thermophilic PET-degrading enzyme — and produced a ranked engineering strategy.
🔥
25 Elite Hotspot Positions Identified
Positions where >70% of amino acid substitutions enhance activity. Positions 83, 50, 10, 183, 20 are top priority targets.
Top mutation: N215H → +1.84× activity
🧊
3 Critical Coldspot Positions Protected
Positions 92, 212, 133 are structurally or catalytically essential — likely the catalytic triad (Ser131-His208-Asp176) region. Must avoid in engineering.
Any substitution → >50% activity loss
🔬
MULTI-evolve Strategy Recommended
Agent identified compatibility with the MULTI-evolve framework (Tran et al., Science 2026): train neural net on pairwise epistasis data → predict optimal 3–7 mutation combinations.
Estimated: 256× improvement potential (cf. APEX2)
// The Platform
LabClaw — 211 Production-Ready Bio Skills
AminoClaw is powered by LabClaw, a modular skill library that teaches the agent when and how to use every frontier biomedical tool.
🧬
Biology & Bioinformatics
66 skills
💊
Drug Discovery
36 skills
🏥
Medical & Clinical
20 skills
📚
Literature & Search
29 skills
🤖
Lab Automation
7 skills
⚙️
Data Science & ML
48 skills
// Market Opportunity
A Market Ready for Disruption
Protein engineering is a cornerstone of drug discovery — and it's overwhelmingly bottlenecked by computational access, not biological creativity.
$10B+
AI drug discovery market by 2030
10M+
Life scientists globally
$2M+
Avg. cost of one protein engineering campaign
<1%
Biologists with AI/ML pipeline access
Biology-first UX — no coding required
End-to-end automation, not point tools
Modular skill layer — works with any frontier model
Proven results: competition-grade outputs
// Get In Touch
Ready to Accelerate
Your Research?
Whether you're designing binders, engineering enzymes, or analyzing DMS data — AminoClaw's agentic pipeline turns your biological intuition into results.
RBX1 Case Study ↗
zczlwl3@hotmail.com