A Framework for Predicting RNA-Ligand Interaction

This is a pipeline developed for screening RNA-ligand interaction. This pipeline combines ab initio RNA structure prediction and geometric deep learning for large-scale RNA-Ligand interaction screen. Our approach predicts RNA 3D structures, trains a geometric deep learning-based scoring model, generates binding complex candidates, and systematically evaluates potential ligands. Uniquely tackling RNA-Ligand interactions in three-dimensional space without experimentally determined crystal structures, our pipeline offers atomic-level assessment and holds promise for advancing RNA-small molecule interaction understanding and RNA-targeted therapeutic design.

RNA 3D structure prediction

Detailed workflow and instructions were described in the structure_prediction directory. Check for details.

Geometric deep learning based scoring model

• Data preparation

  • check out scoring_model/processing_ligand/ for details
  • including
    • parallel RNA-Ligand docking
    • parallel ligand parsing, and
    • data acquisition from PDB

• Model training

  • run scoring_model/run_wligand.sh to train the model in default setting

• Experiments

  • check out scoring_model/experiments/ for details
  • including:
    • method comparison
    • discriminative selection experiment

  Reference

  • Equiformer backbone: Liao, Yi-Lun, and Tess Smidt. "Equiformer: Equivariant graph attention transformer for 3d atomistic graphs." arXiv preprint arXiv:2206.11990 (2022).