name: bio-protac-degraders description: Designs PROTACs, molecular glues, and bivalent degraders with explicit handling of E3 ligase choice (VHL, CRBN, IAP, MDM2, KEAP1), linker design (length, composition, rigidity), ternary complex prediction (PRosettaC, DeepTernary, AlphaFold3), cooperativity (alpha), DC50 / Dmax characterization, hook effect, and prediction-experiment reconciliation. Use when designing targeted protein degraders, planning linker SAR, predicting ternary complex stability, or building generative degrader workflows. origin: openai4s category: bioskills/chemoinformatics metadata: tool_type: python primary_tool: PRosettaC third_party: name: GPTomics/bioSkills repository: https://github.com/GPTomics/bioSkills commit: d91ed3d563019e649dc854c56ccd62551359488a license: MIT
Reference examples tested with: PRosettaC (web service), DeepTernary research code, AlphaFold3, Boltz-1 / Boltz-2, RDKit 2024.09+, OpenMM 8.1+ (for ternary MD).
Before using code patterns, verify installed versions match. If versions differ:
- Python: pip show <package> then help(module.function) to check signatures
If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.
Design bifunctional molecules (PROTACs) that recruit an E3 ubiquitin ligase to a target protein, inducing target ubiquitination and proteasomal degradation. PROTACs differ from traditional drugs: a productive ternary complex (target + PROTAC + E3) is required, not just target binding. The modality has produced clinical programs, but their development and regulatory status changes rapidly and must be checked from current sources. PROTAC design balances target ligand binding, E3 ligand binding, linker geometry (length, rigidity, chemistry), cooperativity, dose-dependent ternary-complex formation, and cell permeability. Negative cooperativity and the high-concentration hook effect are distinct phenomena, although cooperativity can influence the dose-response profile.
For target ligand design, see chemoinformatics/virtual-screening and chemoinformatics/admet-prediction. For linker-only enumeration, see chemoinformatics/reaction-enumeration. For generative linker design, see chemoinformatics/generative-design.
| Recruited UPS component | Ligand series | Published design context | Limitations |
|---|---|---|---|
| VHL | VL-269 (Gechijian et al. 2018) | Published VHL-recruiting degraders | Expression and productive geometry are system-dependent |
| CRBN (cereblon) | thalidomide, pomalidomide | Extensively used recruiter series | Neosubstrate liabilities depend on recruiter and context |
| IAP (XIAP, cIAP1) | SMAC-mimetic-derived recruiters | Published IAP-recruiting degraders | Target scope and cellular effects require validation |
| MDM2 | nutlin-derived recruiters | Published MDM2-recruiting degraders | Target diversity and pathway effects require validation |
| KEAP1 | KEAP1-directed recruiters | CUL3-KEAP1 recruitment studies | Specialized use and limited comparative validation |
| DCAF15 | Aryl sulfonamides such as E7820 | DDB1-CUL4 / DCAF15 systems | Molecular-glue and degrader mechanisms require careful distinction |
| RNF114 | Nimbolide, EN219 | Covalent RNF114 recruitment | Limited tooling |
| RNF4 | CCW16 | Covalent RNF4 recruitment | Limited tooling |
| UBE2D (E2, not E3) | EN450 | Covalent molecular-glue mechanism involving NFKB1 | Do not classify as an E3-ligase recruiter |
Decision: Select an E3 recruiter using evidence for ligand availability, target/E3 geometry, cellular expression, neosubstrate liabilities, and the intended biological system. CRBN and VHL are common starting points with extensive published examples, but neither is a universal first choice.
Linkers tune ternary complex geometry and stability. The ranges below are exploratory starting points, not validated acceptance criteria:
| Property | Range | Effect |
|---|---|---|
| Linker length | Project-defined enumerated series | Critical; geometry-dependent |
| Linker rigidity | Flexible (PEG) vs rigid (piperazine, pyridine) | Changes the accessible conformational ensemble |
| Linker chemistry | PEG, alkyl, piperazine, triazole, ether, amide | PEG common; rigid for tighter binding |
| Click chemistry compatibility | Triazole-forming routes are one option | Requires route- and attachment-specific synthesis review |
| Molecular size and polarity | Measure across the designed series | Permeability, solubility, and exposure depend on the complete molecule and its conformations |
Critical: The "Goldilocks linker length" is target-specific. Too short can create a ternary clash; too long can impose an unfavorable entropic cost or permit unproductive geometries. Enumerate a series around the geometry supported by the binary structures rather than assuming a universal optimal range.
| Goal | E3 / linker | Tools |
|---|---|---|
| Initial degrader series | Compare supported recruiter and linker variants | PRosettaC for ternary hypotheses |
| Reduce recruiter-specific liabilities | Compare alternative E3 recruiters and linker geometries | Structural hypotheses + cellular selectivity validation |
| Target with prior recruiter-specific evidence | Reproduce the supported recruiter context, then vary deliberately | Match the published and intended biological systems |
| Targeted protein degradation program | Select E3 using geometry, expression, and liabilities | Structural and experimental validation track |
| Novel target without an established ternary model | Multiple E3 / linker variants | Combinatorial design + PRosettaC |
| Molecular glue (non-PROTAC) | Use a glue-specific discovery strategy | Distinct mechanism; do not treat as linker design |
| Characterize cooperativity | Structural hypotheses plus experiment | ITC or SPR/BLI with matched binary and ternary measurements |
| Cell-active candidate | Standard development | PK + degradation cellular assays |
| Tool | Approach | Strength | Fails when |
|---|---|---|---|
| PRosettaC | Constrained PatchDock, RosettaDock refinement, PROTAC conformer generation, repacking, and clustering | PROTAC-specific published workflow | Performance varies by complex; Rosetta/service requirements |
| DeepTernary | Equivariant deep learning | Fast; SE(3) | OOD chemistry |
| AlphaFold3 | Unrestrained whole-complex prediction | Accepts proteins and ligands | No arbitrary user distance-restraint interface; benchmark PROTAC use |
| Boltz-1 / Boltz-2 | Unrestrained whole-complex prediction | Open local models | Limited PROTAC-specific validation |
| HADDOCK | Information-driven, restraint-guided docking | Mature integrative docking framework | Manual restraint specification |
Decision: Use a PROTAC-specific method such as PRosettaC for first-pass ternary modeling. AlphaFold3 or Boltz can provide unrestrained whole-complex predictions, but should be benchmarked on relevant ternary complexes. DeepTernary is released research code rather than a hosted API; validate it against relevant structures before prospective ranking.
Cooperativity quantifies how the ternary complex stabilizes (or destabilizes) the binary binding:
alpha = (Kd_binary,target) / (Kd_ternary,target)
Positive cooperativity can favor ternary-complex formation, but the preferred alpha is system- and assay-dependent and does not alone establish degradation efficacy. Alpha must be measured from binary and ternary binding experiments; a predicted structure does not directly provide it.
Measure with ITC (isothermal titration calorimetry) or SPR/BLI titrations of binary vs ternary.
In cellular assays: - DC50: PROTAC concentration for 50% degradation (analogous to IC50) - Dmax: maximum fraction degraded at any concentration
| Property | What to report | Interpretation |
|---|---|---|
| DC50 | Concentration producing 50% of the assay's fitted maximal degradation | Compare only across matched assay conditions; no universal clinical cutoff |
| Dmax | Maximum observed or fitted degradation and uncertainty | Required depletion is target- and phenotype-dependent |
| Hook effect | Full concentration-response range and concentration of any downturn | A high-concentration effect; its location is system- and assay-dependent |
| Cooperativity | Alpha from matched binary and ternary binding experiments | Distinct from the hook effect and insufficient by itself to predict degradation |
Hook effect: at high PROTAC concentrations, binary complexes (PROTAC-target alone, PROTAC-E3 alone) dominate, and ternary complex formation drops. Dose-response curves are bell-shaped.
Goal: Predict 3D structure of target-PROTAC-E3 ternary complex.
Approach: 1. Start with binary co-crystals: target + target-ligand pose; E3 + E3-ligand pose 2. Connect via linker enumeration (combinatorial) 3. Score by geometric feasibility (linker length, no clashes) 4. Refine with energy minimization
# Pseudo-code workflow
def predict_ternary(target_pdb, target_ligand_sdf,
e3_pdb, e3_ligand_sdf, linker_smiles):
# 1. Place binary complexes in same coordinate frame
# 2. Enumerate linker connectivity from target-ligand exit vector to e3-ligand entry vector
# 3. Score by total linker length, RMSD to expected geometry
# 4. Apply a documented refinement protocol and test convergence
return ternary_poses
For a production workflow, use PRosettaC or provide both proteins and the complete PROTAC as components of an unrestrained AlphaFold3 input. AlphaFold3 does not expose arbitrary chain-chain distance restraints; compare predicted interfaces and confidence with known complexes or a PROTAC-specific method.
from rdkit import Chem
from rdkit.Chem import AllChem
def attachment_distance(target_ligand, e3_ligand,
target_attachment_idx, e3_attachment_idx):
"""
Measure an attachment-point distance after both ligands have been placed in
the same ternary-complex coordinate frame.
"""
p1 = target_ligand.GetConformer().GetAtomPosition(target_attachment_idx)
p2 = e3_ligand.GetConformer().GetAtomPosition(e3_attachment_idx)
return p1.Distance(p2)
An attachment-point distance does not map uniquely to a linker atom count: bond geometry, rigidity, branching, solvation, and the relative protein orientation all matter. Enumerate chemically synthesizable linker candidates, sample their conformers in the ternary geometry, and retain candidates that can connect without severe strain or clashes.
REINVENT 4 can generate linkers, but it does not provide the ternary_score / deepternary interface shown in some informal examples. Export generated candidates, run an installed and validated ternary-prediction workflow separately, and then join the structural scores back to the candidates. Do not assume DeepTernary is a web API or a built-in REINVENT scoring component.
Trigger: Target's known binding mode incompatible with E3 ligase orientation.
Mechanism: The selected binary poses, exit vectors, linker conformations, or protein orientation may not support a compatible ternary geometry.
Symptom: Low ternary complex scores; high RMSD across replicates.
Fix: Try a different E3 such as CRBN or VHL and compare against experimentally resolved ternary complexes with compatible exit-vector geometry.
Trigger: Target ligand or E3 ligand outside training distribution.
Mechanism: A ligand, linker, target, or E3 outside the method's validated domain may require extrapolation.
Symptom: Predicted ternary complex unrealistic.
Fix: Compare with an independently configured structural method and relevant known complexes; validate prospective ranking experimentally.
Trigger: PROTAC concentration becomes high enough that separate target-PROTAC and E3-PROTAC binary complexes compete with productive ternary-complex formation.
Mechanism: Saturation by binary complexes reduces the population of productive ternary complex. Negative cooperativity can worsen ternary formation but is not the definition of the hook effect.
Symptom: Degradation increases and then decreases across a sufficiently broad concentration-response experiment.
Fix: Confirm the downturn experimentally over a broad dose range, then optimize ternary-complex geometry, cooperativity, exposure, and dosing without assuming that linker shortening alone will solve it.
Trigger: Measured permeability or cellular exposure is poor relative to biochemical activity.
Mechanism: Size, exposed polarity, conformation, ionization, or efflux may limit intracellular exposure.
Symptom: Cellular degradation potency is substantially worse than biochemical ternary-complex or binding measurements.
Fix: Optimize linker and exposed polarity using measured permeability, solubility, and intracellular exposure across the series. Do not impose a universal MW or TPSA cutoff.
Trigger: Computing linker length from binary models without ternary refinement.
Mechanism: A distance from separately positioned binary structures does not determine the accessible ternary geometry or linker conformational ensemble.
Symptom: PROTACs synthesized at wrong linker length; no degradation.
Fix: Use a ternary structural hypothesis to define a chemically diverse linker series, then compare conformational feasibility and experimental degradation across that series.
Trigger: Designing as PROTAC when target lacks defined ligand.
Mechanism: A molecular glue stabilizes or induces a protein-protein interaction without the two-ligand-plus-linker architecture assumed by a PROTAC workflow.
Symptom: Design too rigid; no degradation despite ternary prediction.
Fix: For targets without known ligand, consider molecular glue discovery instead.
| Aspect | PRosettaC | AlphaFold3 |
|---|---|---|
| Approach | PROTAC-specific Rosetta sampling | Unrestrained foundation-model prediction |
| Accuracy | Higher average DockQ in one 36-structure comparison, but only 25 complexes were modeled and most predictions were low quality | Limited PROTAC-specific validation |
| Speed | Measure for the installed workflow and hardware | Measure for the selected service or local hardware |
| Access | Web service | AlphaFold Server or local installation, subject to their terms and limits |
| Restraints | Method-specific setup | No arbitrary user distance restraints |
| Decision | Use as a PROTAC-specific structural hypothesis | Use as an independently benchmarked structural hypothesis |
Use PRosettaC or another benchmarked structural method to generate hypotheses, then measure ternary binding/cooperativity and cellular degradation experimentally. Do not treat any one modeling method as a validated universal ranker.
| Symptom | Cause | Fix |
|---|---|---|
| PRosettaC fails to converge | Input geometry, sampling, or service/configuration problem | Inspect inputs and logs; compare justified recruiter/linker hypotheses |
| DeepTernary returns clashing pose | Prediction outside a validated domain or incorrect interface | Inspect confidence and clashes; compare an independent method and known structures |
| AlphaFold3 ternary unrealistic | Unrestrained prediction has a low-confidence or incorrect interface | Inspect confidence and compare with PRosettaC or known ternary structures |
| Cellular phenotype disagrees with target-degradation assays | Exposure, off-target degradation, assay timing, or pathway effects | Measure target engagement/degradation and use proteome-wide selectivity assays where appropriate |
| Degradation decreases at high PROTAC concentration | Hook effect from competing binary complexes | Confirm with a broad dose range; optimize ternary geometry and exposure |
| Synthesis is impractical | Proposed connectivity lacks a credible route | Obtain medicinal-chemistry review and redesign attachment chemistry or linker |
| Poor permeability or intracellular exposure | Size, exposed polarity, conformation, or efflux | Measure the bottleneck and optimize the series; avoid a universal size cutoff |