1,567 AI-Designed GID4 Binders: An Open Dataset for Targeted Protein Degradation

We've released an open dataset of 1,567 de novo, generative-AI–designed small-molecule binders of human GID4, each provided as a docked protein–ligand complex — generated by the Technetium GA-II pocket-conditioned platform and released under CC-BY-4.0.

Why GID4?

GID4 is the substrate-recognition subunit of the human CTLH E3 ubiquitin-ligase complex and an emerging handle for targeted protein degradation (TPD) and molecular glues. Receptor coordinates are based on PDB 7SLZ.

What's in the dataset

• 1,567 self-contained *_complex.pdb files (receptor + 3D docked ligand pose + a 2D↔3D atom map), 1,497 unique ligand SMILES.
• A loadable Parquet table with canonical SMILES and RDKit descriptors (MW, cLogP, TPSA, Fsp3, HBD/HBA, rotatable bonds, ring count) — powering an in-browser Dataset Viewer.
• The library sits in Lipinski/Veber-friendly space (median MW 420, cLogP 2.4, TPSA 85) with notable 3D character (median Fsp3 0.36) — appropriate for GID4's shallow, partly peptidic pocket.

Load it in one line

from datasets import load_dataset

ds = load_dataset("Tc-43/GID4_7SLZ_20260620", split="train")
df = ds.to_pandas()   # compound_id, smiles, complex_file, descriptors ...
The full 3D poses are in GID4_7SLZ_20260620.zip, joinable on complex_file / compound_id.

Explore the poses in 3D
We also built an interactive viewer — browse a representative sample of the binders docked
in the GID4 pocket, rotate/zoom, and inspect each design:

👉 Space: https://huggingface.co/spaces/Tc-43/gid4-7slz-explorer

Links
📦 Dataset: https://huggingface.co/datasets/Tc-43/GID4_7SLZ_20260620
🗂️ All 9 design sets (oncology, immunology, TPD): https://huggingface.co/collections/Tc-43/technetium-ga-ii-generative-design-sets-6a3c0385ed11bfca1fdcfb07
🧬 Receptor: PDB 7SLZ
Provenance & intended use
These are computationally generated designs and docked poses — not experimentally
validated binders; no claim of activity is made. They're intended for machine-learning,
cheminformatics, generative-model benchmarking, and docking-pose research on a well-defined
TPD target. Feedback and collaborations welcome.