Machine learning
The advantages of computational tools
One of the major challenges in improving medical devices and tissue engineering strategies is understanding the exact interaction between the biomaterial and the human body. Computational tools can help to study the multiscale, spatiotemporal complexities at both sides of the interface – the material on the one hand and the organism on the other hand – in a quantitative way. Moreover, computational models can screen for promising hypotheses, predict variables inaccessible for measurements and inform the experimental design to reduce the amount of in vitro experiments (as well as the associated time and costs). Our research focuses on a suite of modelling techniques ranging from mechanistic (hypothesis-based) models to empirical (data-driven) models and covering the intracellular and cellular scale. Each model system has its own benefits and limitations which determine the application for which it can be used. More specifically, we have established an automatic imaging pipeline to process the (high-content) images obtained from the TopoChip screens. Further, we used Random Forest classification algorithms to assess the quantitative material activity relationship between surfaces classes based on expression levels of different markers such as ALP or Oct4 and topographical design. In addition, we used supervised machine learning analysis, implemented in Cell Profiler Analyst to identify cell shape clusters. In the next step, we employ advanced machine learning techniques, including Deep Convolutional Neural Networks to derive predictive models of cell-surface associations (paper in preparation). Also, we developed topographical surface optimization approach based on genetic algorithms (paper in preparation).
Further reading
Tensorflow
M. Hulsman et al. ActaBiomater, 2015
Reimer et al. Sci. Rep., 2016
A. Vasilevich et al. Trends Biotechnol.,2017
F. Hulshof et al. Biomaterials, 2017
N. Beijer et al. Advanced Biosystems, 2017