value
| - Computational prediction of bioactivity has become a critical aspect of modern drug discovery as it mitigates the cost, time, and resources required to find and screen new compounds. Deep Neural Networks (DNN) have recently shown excellent performance in modeling Protein-Ligand Interaction (PLI). However, DNNs are only effective when physically sound descriptions of ligands and proteins are fed into the neural network for processing. Furthermore, previous research has not incorporated the secondary structure of proteins in a meaningful manner. In this work, we developed a DNN framework that utilizes the secondary structure information of proteins which is extracted as the curvature and torsion of the protein backbone to predict PLI. We demonstrate how our model outperforms previous machine and non-machine learning models on four major datasets: humans, C.elegans, DUD-E, and BindingDB. Visualization of the intermerdiate layers of our model shows a potential latent space for proteins which extracts important information about the bioactivity. We further investigate the inner workings of our model by visualizing the most important aspects in a protein that the model finds influential. We observed that our model learns important information about possible locations where a ligand would bind including binding sites, allosteric sites and cryptic sites, regardless of the conformation used. This work opens the door to the exploration of secondary structure based deep learning in general, which is not just confined to protein-ligand interactions.
|