PROTAC and Bivalent Degrader Design Usage Guide

Overview

Design PROTACs (bivalent molecules recruiting E3 ligase to target for proteasomal degradation). Balance target-ligand, E3-ligand, linker geometry, cooperativity, and cell permeability. Cover ternary complex prediction (PRosettaC, DeepTernary, AlphaFold3), cooperativity, hook effect, and DC50/Dmax.

Prerequisites

pip install rdkit
# PRosettaC: web service (prosettac.weizmann.ac.il)
# DeepTernary: GitHub installation
# AlphaFold3: use AlphaFold Server or a licensed local installation; arbitrary distance restraints are not supported

Quick Start

Tell the AI agent what to do: - "Design PROTAC for kinase X target using CRBN E3" - "Enumerate linkers between target-ligand and VHL ligand" - "Predict ternary complex for kinase + PROTAC + CRBN" - "Optimize linker length to maximize cooperativity"

Example Prompts

CRBN PROTAC design

"For target kinase X (PDB 5XYZ), design an exploratory PROTAC series using a justified E3 recruiter. Vary linker composition and geometry around candidates compatible with the binary structures. Output connected, synthesis-review-ready SMILES with structural scores reported as hypotheses."

Ternary complex prediction

"Predict the ternary structure for this target-ligand-E3 PROTAC complex using PRosettaC. Report structural and interface scores; state that cooperativity alpha requires a binding experiment."

Linker optimization

"Given target-ligand and E3-ligand exit vectors in a shared structural hypothesis, suggest a small linker series spanning rigid and flexible chemistries. Explain how conformer feasibility will be checked rather than inferring atom count from distance alone."

VHL alternative design

"Switch from CRBN to VHL E3. Adjust linker to maintain ternary geometry. Re-predict ternary complex."

What the Agent Will Do

1. Define target ligand (from co-crystal or docked) and E3 ligand (pomalidomide / VHL ligand).
2. Compute distance between attachment points on each ligand.
3. Enumerate a linker series around geometries supported by the binary structures.
4. Combine target-linker-E3 SMILES.
5. Predict ternary complex via PRosettaC (or DeepTernary).
6. Score by linker geometry and structural/interface metrics; report experimental cooperativity separately when available.

Tips

• CRBN and VHL have extensively published recruiter series, but choose the E3 using target/E3 geometry, expression in the intended system, ligand availability, and known liabilities.
• Optimal linker length is target-specific; use structural hypotheses to choose an exploratory series rather than a universal range.
• Positive cooperativity may be beneficial, but alpha is assay- and system-dependent and must be measured experimentally.
• Hook effect at high concentration; bell-shaped dose-response.
• Permeability can be a bottleneck; compare measured permeability and cellular activity across the series rather than applying a universal MW or TPSA cutoff.
• The linker-enumeration example requires each explicit dummy attachment to use a single bond; reject other dummy-bond orders unless the intended connection chemistry has a separately validated bond-order rule.

Related Skills

• chemoinformatics/molecular-io - Parse ligand SMILES
• chemoinformatics/reaction-enumeration - Linker enumeration
• chemoinformatics/generative-design - REINVENT linker mode
• chemoinformatics/conformer-generation - Ternary conformer sampling
• chemoinformatics/virtual-screening - Validate target ligand binding
• chemoinformatics/free-energy-calculations - Ternary ABFE
• chemoinformatics/admet-prediction - PROTAC ADMET specifics
• structural-biology/structure-io - PDB / mmCIF for ternary complex