Computational Prediction of Intracellular Signaling Behavior via Machine Learning
Résumé
Assessing the dynamics of intracellular signaling processes under various conditions such as protein-protein interactions, protein conformational changes, dose-dependent drug effects, and ligand binding is crucial for biologists and pharmacologists. However, generating such data can be time-consuming and costly. In this study, we used data obtained by Bioluminescence Resonance Energy Transfer (BRET) in live cells to develop models to predict the temporal dynamics of intracellular second messenger cAMP (cyclic adenosine monophosphate). We formulated the task of predicting time series in a supervised manner and employed decision tree, random forest, and XGBoost models within a multiple-input, multiple-output framework, enabling simultaneous forecasting of multiple future time steps. Our findings demonstrate that our model achieves, in the main experiment, a mean absolute error of less than 0.018 on the testing set. This model is computationally efficient, requiring only ≈ 16 seconds for training, validation, and testing. Finally, we identified that BRET signal intensities are the most important feature for predictions among the set of features in the models. These results highlight the potential of machine learning models for advancing research in dynamical biological processes, particularly in cellular signaling and pharmacological systems.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |