Deep Docking-Based Virtual Screening for NirA Inhibitors

This repository implements a deep learning-guided virtual screening protocol for identifying potential inhibitors of the NirA protein, using an ultra-large chemical library. The workflow is based on the original Deep Docking framework by James Gleave, modified and optimized for phytochemical screening against NirA, a key enzyme of interest in this study.

Overview of the Protocol

The methodology combines deep neural network-based filtering with molecular docking to rapidly prioritize promising binders from a massive compound library.

Key Steps:

1. Sampling from Library: Compounds are sampled from public databases (ZINC, DUDE, IMPPAT, COCONUT).
2. Data Preparation: Decoys and actives are prepared based on molecular weight and activity criteria.
3. Dataset Split: ~9 million phytochemicals are split into:
   • Training
   • Validation
   • Testing
4. Deep Neural Network Training:
   • Hyperparameter tuning
   • Training set size optimization
   • Model selection
5. Virtual Screening:
   • Predict docking scores
   • Discard low-score compounds
   • Retain top hits as virtual hits
6. Validation: Top hits are validated using molecular docking against the NirA protein.
7. Iterative Refinement: Top compounds are re-scored, re-trained, and filtered in successive rounds.

Results Summary

• Library Size: ~9 million

• Top Scoring Hits: 44,168 compounds

• Databases: ZINC, DUDE, IMPPAT, COCONUT

• Architecture: Feed-forward Deep Neural Network with hyperparameter tuning and iterative refinement

• Reference

If you use this repository or base your work on it, please cite:

• Ton et al. (2020) Deep Docking: A Deep Learning Platform for Augmentation of Structure Based Drug Discovery DOI

• Original Deep Docking GitHub

• Contact

For queries, contact: thedrsoham[at]gmail[dot]com or open an issue in this repository.

"Accelerating ultra-large-scale docking with deep learning for NirA-targeted compound discovery."