Decomposing Deep Learning Predictions to Uncover Beta-2 Adrenergic Conformational Landscapes

Understanding the mechanisms underlying protein receptor function and their roles in cellular processes is crucial for advancing biological discovery. In this study, we employ a one-dimensional convolutional neural network to classify receptor conformations from molecular dynamics simulations of a G-protein-coupled receptor, the β<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf>-adrenergic subtype: a key drug target known in the treatment of various human conditions. In order to address the complexity and subtlety of classifying conformational states that influence the receptor response, we integrated a decomposition-based explainability method inspired by ANOVA techniques. This approach allows us to break down the predictions of the black-box model and highlight the key features driving classification decisions. Equally important to model explanations is the rigorous preprocessing of the MD simulation data, which ensures precise model learning and thus accurate explanations. This approach unravels the predictions of the model, offering insight into the dynamics of receptors and enhancing their interpretability. The results underscore the effectiveness of combining deep learning with advanced decomposition techniques to achieve not only accurate classification, but also a transparent view of the decision-making process.