PharmRL: A New Method for Designing Pharmacophores Using Deep Learning
Author Information
Author(s): Rishal Aggarwal, David Koes
Primary Institution: University of Pittsburgh
Hypothesis
Can a deep learning method effectively identify pharmacophores in the absence of a ligand?
Conclusion
PharmRL successfully generates functional pharmacophores even when cognate ligands are unavailable.
Supporting Evidence
- PharmRL shows better prospective virtual screening performance on the DUD-E dataset than random selection.
- Experiments on the LIT-PCBA dataset demonstrate efficient identification of active molecules.
- PharmRL effectively identifies prospective lead molecules in the COVID moonshot dataset.
Takeaway
This study created a computer program that helps find important parts of molecules that can help design new drugs, even when the usual reference molecules aren't available.
Methodology
A convolutional neural network (CNN) was trained to identify pharmacophore features, and a deep geometric Q-learning algorithm was used to select optimal subsets of these features.
Potential Biases
The pharmacophore features may be influenced by biases in the docking and simulation protocols used.
Limitations
The method relies on the quality of the training data and may not perform well in all scenarios due to inherent biases in the dataset.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website