|
|
Despite the increasing interest in psychedelics for therapeutic applications, our understanding of how these substances interact with the 5HT2A receptor remains limited. A deeper understanding of these interactions is essential for unlocking the full potential of psychedelics in targeted treatments. In this study, we present a novel geometrical deep learning framework that leverages graph convolutional neural networks (GCNs) to systematically analyze the 3D conformations of the 5HT2A receptor, as captured through molecular dynamics (MD) simulations.
Our framework is trained on graph representations of MD simulations featuring the 5HT2A receptor bound to various ligands. The models predict the bound ligand based solely on the receptor’s structure, learning to recognize important patterns in the 5HT2A receptor. Once trained, we use various explainability mechanisms to identify the receptor regions crucial for accurate predictions.
The deep-learning approach we employ uncovers key receptor regions, including the toggle switch, NPxxY motif, and side-extended cavity, which have been highlighted in the literature and confirmed by experimental studies as critical components in 5HT2A receptor dynamics. By offering a more accessible and comprehensive understanding of the molecular basis of the functional selectivity observed between different psychedelics and other ligands, the framework can guide the development of more effective and targeted therapeutics. |
