|
Project properties |
|
| Title | Exploring the use of neural differential equation models for experimental design and parameter estimation |
| Group | Systems and Synthetic Biology |
| Project type | thesis |
| Credits | 36 |
| Supervisor(s) | Rob Smith |
| Examiner(s) | robert1.smith@wur.nl |
| Contact info | robert1.smith@wur.nl |
| Begin date | 2025/08/20 |
| End date | |
| Description | At SSB, our projects are either offered as computational or experimental projects. Nearly all projects can be altered for either BSc or MSc students (as required). For more information, please contact Rob Smith (robert1.smith@wur.nl). Note that we will evaluate applications based on discussing the match between a student’s current competencies and the competencies needed for the project – sending a CV is not required.
This thesis project will be a computational project. The project will be supervised by Rob Smith. Over recent years, the application of neural networks to aid development of mathematical models has increased (Noordijk et al. 2024). This includes for estimating missing parts of a model (often referred to as surrogate models), responses of systems to noise, enhanced data fitting and parameter estimation, and to estimate how a model’s dynamics would change in response to an altered input value (Su et al. 2022, Giampiccolo et al. 2024, van Aalst et al. 2025, van Tegelen et al. 2025). This suggests that neural ODEs could, in the future, be used to speed up the design of experiments to optimise a system’s response to experimental inputs after data-fitting. An example of how model-based design is traditionally achieved with ordinary differential equation (ODE) models can be seen in Martin-Pascual et al. (2024). In this work, ODEs were fit to experimental data of curcumin production in the presence of different enzymes. The model was then used to accurately predict which enzymes needed to be overexpressed to increase curcumin levels. This provides scope to test neural ODE performance: Can neural ODEs improve data fitting relative to the ODE model? Can neural ODEs predict dynamics missing from the ODE model? And, can neural ODEs predict enzyme levels needed to maximise curcumin production? Using both real and simulated datasets, we can explore the role of neural ODEs for experimental design with an eye on speeding up this process in the future. References related to project: Su et al. (2022) doi: 10.48550/arXiv.2209.01862 Giampiccolo et al. (2024) doi: 10.1038/s41540-024-00460-3 Martin-Pascual et al. (2024) doi: 10.1101/2024.02.08.579459 Noordijk et al. (2024) doi: 10.3389/fsysb.2024.1407994 van Aalst et al. (2025) doi: 10.1101/2025.05.06.652335 van Tegelen et al. (2025) doi: 10.48550/arXiv.2507.19036 |
| Used skills | |
| Requirements | Interested students should be comfortable with mathematical modelling using ordinary differential equations and programming in Python. Knowledge of neural networks or machine learning is also useful. |