CoolFace
Apppublic

andysenner/aptamergnn

sourceHugging Facemitupdated 4mo agoView on Hugging Face
0likes
App README

AptamerGNN

Structure-based aptamer binding affinity prediction using Graph Neural Networks.

Upload a protein-aptamer complex PDB file and receive a predicted binding affinity (Kd) with 3D pocket visualization.

How it works

  1. 1.Upload a PDB file containing a protein-aptamer complex
  2. 2.The pipeline identifies the binding pocket (8 Å cutoff)
  3. 3.A GATv2 graph neural network predicts pKd from atom-level structural and physicochemical features
  4. 4.Results include predicted Kd, confidence, pocket residues, and 3D visualization

Model

  • —Architecture: GATv2 (3-layer, 64 hidden dim, 4 attention heads)
  • —Training data: 721 protein-aptamer complexes from PDBbind
  • —Node features: 98 dims (element, atom name, residue identity, hydrophobicity, charge, polarity, aromaticity, H-bond properties, occupancy, B-factor)
  • —Edge features: distance, direction vector, interface flag
  • —Test set RMSE: ±1.24 pKd units
  • —Test set Pearson R: 0.55

Disclaimer

This tool is provided for research use only. Predictions have inherent uncertainty (±1.24 pKd units, ~1 order of magnitude in Kd) and require experimental validation before any clinical or therapeutic application.

The authors disclaim all liability for outcomes resulting from use of this tool. Users are responsible for ensuring their use complies with applicable laws, institutional policies, and ethical standards.

License

MIT License — see LICENSE file.

Citation

If this tool contributes to your research, please cite:

AptamerGNN: Structure-based aptamer binding affinity prediction.
[Andrew Senner, 2026]. https://huggingface.co/spaces/[your-username]/aptamergnn