HybridML: Open source platform for hybrid modeling
Combining data-driven models with mechanistic model components reduces the data demand and enables extrapolation. HybridML is an open-source platform for building, training, and evaluating such hybrid models.
Building, training, and evaluating hybrid models is cumbersome with standard frameworks. HybridML is an open-source modeling platform in which hybrid models, i.e., combinations of artificial neural networks, arithmetic expressions, and differential equations, can be trained. It uses TensorFlow for training the neural networks and CasADi to integrate ordinary differential equations and to provide gradients of the differential model equations, which enables continuous-time representations. HybridML also provides a JSON interface for model development.
HybridML was applied to an industrial case study, in which the trained model predicts drug concentrations over time based on physiological information about the patients. To demonstrate its versatility, it was also used to model the spread of the COVID-19 pandemic in German federal states based on the states' socio-economic attributes.
For our current work on hybrid models, see the project page Hybrid modeling.
Open-source code
The open-source code of HybridML is available on GitHub.
Applications
HybridML was used to predict the concentration of the drug fluvoxamine in patients' blood over time. The input to the model is a set of covariates (attributes) such as gender, weight, and age of each patient, and the initial dose of fluvoxamine given to that patient. Measurements of the fluvoxamine concentration in the patients' blood over time and the time since the patient took the medicine are used to estimate the model parameters.
Key publications
- Merkelbach, K., Schweidtmann, A. M., Müller, Y., Schwoebel, P., Mhamdi, A., Mitsos, A., Schuppert, A., Mrziglod, T., & Schneckener, S. (2022). HybridML: Open source platform for hybrid modeling. Computers & Chemical Engineering, 160, 107736. doi:10.1016/j.compchemeng.2022.107736 · Details