---
license: cc-by-4.0
pretty_name: VHL Non-Proline Binder Designs (PDB 9GIO)
task_categories:
- tabular-regression
- graph-ml
- other
size_categories:
- n<1K
configs:
- config_name: default
  data_files:
  - split: train
    path: designs.csv
language:
- en
tags:
- chemistry
- biology
- drug-discovery
- molecular-docking
- protac
- targeted-protein-degradation
- vhl
- generative-design
- von-Hippel-Lindau
- VCB
- elongin-B
- elongin-C
- E3-ligase
- E3-ubiquitin-ligase
- ligase-handle
- TPD
- degrader
- non-proline
- hydroxyproline
- HIF-1alpha
- 9GIO
- protein-ligand-complex
- structure-based-drug-design
- de-novo-design
- generative-ai
- ai-generated
- synthetic-data
- small-molecule
- SMILES
- SDF
- cheminformatics
- computational-chemistry
- virtual-screening
- genetic-algorithm
- GA-II
- autodock-vina
- TC-43
- technetium
---


# VHL Non-Proline Binder Designs (9GIO)

279 generatively designed small molecules docked into the **VHL substrate-recognition
pocket** of the VHL–ElonginC–ElonginB (VCB) complex, using **PDB 9GIO** as the receptor.

Molecules were generated by the Technetium **`TC-43.ai`** engine (GA-II generation).
Target selection, structural analysis and dataset curation were performed with **Claude Code**.

## Why these designs matter for PROTAC development

Essentially every VHL-recruiting PROTAC in the literature is built on the same
**(4R)-hydroxyproline** warhead — VH032 and its close analogues. That warhead is an
excellent binder but an expensive one to carry: it contributes a large polar surface
area to a molecule that is already far outside Rule-of-5 space, and PROTAC
**permeability is TPSA- and HBD-limited**.

These designs are a different chemical proposition. **278 of 279 contain no
4-hydroxypyrrolidine and none contain a prolinamide** — they occupy the same pocket
through genuinely non-proline chemotypes (252 distinct Murcko scaffolds across 274
unique molecules).

### Lower TPSA than the proline-based standard

| | TPSA (Ų) | MW | HBD | rot. bonds |
|---|---|---|---|---|
| **These designs** (median) | **89.6** | 450.6 | **2** | **4** |
| VH032 / `3JF` — proline warhead in 9GIO | 111.6 | 472.6 | 3 | 6 |
| `87A` — VHL PROTAC (proline-based) | 208.6 | 936.1 | 5 | 16 |
| `Q43` — VHL PROTAC (proline-based) | 209.4 | 1060.2 | 6 | 15 |

- **227 of 279 (81%) fall below VH032's TPSA**, interquartile range 72.0–107.9 Ų.
- Median TPSA is **22 Ų lower** than VH032, with **one fewer H-bond donor** and
  **two fewer rotatable bonds**.

Because the warhead's polarity is carried directly into the assembled degrader, a
~22 Ų TPSA reduction and the loss of an H-bond donor propagate into the full PROTAC —
the property axis on which proline-based VHL degraders such as `87A` and `Q43`
(TPSA ≈ 209 Ų) are most constrained. Lower donor count and reduced flexibility also
ease the conformational and desolvation penalties that dominate bRo5 permeability.

### They bind where VH032 binds — by a different, more robust mechanism

Superposed on 9GIO, the designed ligands overlap the VH032 site to within **0.13 Å**
(closest approach), and sit **19.8 Å** away from the covalent C77 site also present in
that structure. But they hold that site differently. Running the same interaction
analysis on VH032 in 9GIO and on all 279 designs:

| | VH032 (`3JF`, proline-based) | these designs |
|---|---|---|
| H-bonds | 3, **all sidechain**: Ser111 OG (2.61 Å), His115 ND1 (2.67 Å), Tyr98 OH (2.73 Å) | 2.0 per design, **97% anchored on the His110 backbone** |
| π-stacking | **none** | **95% of designs** — Tyr112, Phe76, Tyr98, Trp88, Phe91, His115 |
| hydrophobic | 5 | 5.8 per design |

VH032 binds as a hydroxyproline mimic of hydroxylated HIF-1α Pro564, and its affinity
rests on three **sidechain** hydrogen bonds. Sidechain donors and acceptors are
rotatable and entropically expensive, and one of the three is geometrically marginal in
this structure (His115, geometry score 0.08).

These designs instead converge on a **backbone** hydrogen bond to His110 — present in
**271 of 279** — combined with **aromatic stacking against a pocket that is unusually
rich in it** (Tyr98, Tyr112, Trp88, Trp117, Phe76, Phe91). Backbone carbonyl geometry is
fixed by the fold rather than by a rotatable sidechain, and stacking carries no
desolvation penalty. Notably, VH032 makes **no π-stacking at all** — the designs exploit
a source of binding energy the substrate-mimetic warhead leaves on the table.

Only 4 of 279 designs hydrogen-bond Ser111. That is by construction, not by omission:
**reproducing the hydroxyproline–Ser111 contact is not required** once a ligand has a
conserved backbone anchor and good stacking, and abandoning it is precisely what frees
these scaffolds from the proline core and its polar-surface cost.

Pocket contacts (heavy-atom, 4.5 Å) across all 279 designs:

| residue | engaged |
|---|---|
| Tyr98 | 100% |
| His110 | 100% |
| Tyr112 | 99% |
| Trp117 | 99% |
| Trp88 | 94% |
| His115 | 78% |

### Every design presents a linker vector

All 279 carry at least one well solvent-exposed heavy atom (≥20 Ų SASA in the complex;
median 6 such atoms per molecule, most-exposed atom median 43.5 Ų), so each offers a
defined attachment point for linker growth toward the E3–target interface.

## Honest limitations

These are **computational designs, not measured binders.** Please read them as
hypotheses:

- **No experimental data.** No binding affinity, no ternary complex formation, no
  degradation. Docking scores (AutoDock Vina, −13.5 to −10.0 kcal/mol) rank poses;
  they are not affinities and should not be read as such.
- **A different binding mode is an untested binding mode.** The His110-backbone-plus-
  stacking anchor set is mechanistically sound and more geometrically reliable than
  VH032's three sidechain hydrogen bonds, but whether it delivers comparable *affinity*
  has not been measured. That is the open question in this set.
- **The TPSA gain is partly traded for lipophilicity.** Median cLogP is 3.4 against
  VH032's 2.2, and median QED is 0.4 against 0.6. Solubility and off-target risk should
  be assessed alongside permeability.
- **Rigid, aromatic-rich.** Median 6 rings vs VH032's 3 — consistent with a
  stacking-driven binding mode, but worth watching for planarity and solubility.
- **Single rigid receptor.** All poses were docked into one 9GIO conformation with no
  induced fit; VHL's pocket accommodates ligands with some plasticity.

## Contents

| file | description |
|---|---|
| `designs.csv` | one row per design — SMILES, docking score, ligand efficiency, computed properties including `tpsa_vs_VH032`, and per-design interaction counts (`his110_backbone_hbond`, `n_pi_stacking`, `best_pi_score`, `ser111_hbond`, …) so every claim above is checkable |
| `ligands.sdf` | 279 ligands with 3D docked coordinates, bond orders and formal charges; all verified against their source SMILES |
| `receptor_9GIO.pdb` | the shared VHL–EloC–EloB receptor |
| `structures/` | 279 full complex PDBs; `REMARK` records carry ligand ID, Vina score and SMILES |

## Reference structure

**9GIO** — *Crystal structure of the VHL–EloC–EloB complex with a covalent compound bound
to C77 of VHL.* Contains both `3JF` (VH032, the proline-based VHL ligand) and `A1IMD`
(the covalent C77 binder). These designs target the `3JF` site.

## Citation

Designs generated by the Technetium `TC-43.ai` engine. Analysis and curation by Claude Code.
Released under CC-BY-4.0.