Bio-HhOST aims to implement “hybrid organoids”, which means systems composed of biological and artificial cells. The purpose of hybrid organoids is to provide a tissue–like an environment for studying biological, e.g., tumours and cells. In this example, tumour cells would be an instance of biological cells, and the environment would be constructed from artificial cells. The rationale for the construction of such a tissue–like system is that neither isolated (tumour) cells nor (tumour) cells embedded in biological tissue are well-suited for reproducible experimental work:

• On the one hand, the behaviour of isolated cells can differ considerably from that of cells embedded in their native tissue, even when considering the same cell type.
• On the other hand, biological tissues are even more complex to characterize and standardize than isolated cells. Studying cells in biological tissue increases uncertainties, the number of known and unknown unknowns, and the necessary experimental effort.

In one sentence, Bio-HhOST aspires to replace the “tissue” with some artificial soft–matter structures (e.g. vesicles, that are liquid volumes enclosed by a lipid bilayer), which are hoped to mimic biological tissue cells but are much simpler than those and produced under reproducible laboratory conditions.

Our partner institutions in Cardiff and Trento will construct these organoids. These research groups build up first on their expertise in soft matter engineering and microfluidics and second on the results we obtained in an earlier joint project (ACDC). The task of ZHAW comprises:

• We assume containers and their connections as given. In silico, we study the dynamics of the chemistry in this spatially heterogeneous setting. Various types of questions are answered:

• Prediction of chemical dynamics: This is the ultimate goal. However, achieving it requires detailed knowledge of reaction rates and transport parameters. Given that the reactions under consideration happen in soft matter systems, these parameters may not be precisely known.
• Classification of behaviour: If parameters are not known, a classification of the possible behaviour is established by analyzing relevant parts of the parameter space.
• Robustness: The robustness of these behaviours is analyzed: Does a small change in the parameter values lead to a completely different behaviour or do we have a robustness that offers the realization of a technically viable system?
• Programmability: Some of these parameters are easy to control, at least in relative terms (doubling the number of pores may lead to doubling transport rates, at least to the first order). What diversity in functional behaviour (“programmability”, to use a more aspirational term) can be achieved by the variation of those processes we know that we can vary them easily?  

• Based on physical dynamics and statistical mechanics, ZHAW supports the design and control of the container assembly process. This includes a variety of tasks:
• Physical assembly: The assembly dynamics of containers is given by the mechanics of the soft matter they are composed of and the hydromechanics of the environment. We use code written during ACDC and standard software such as GROMACS.
• Linking dynamics: The assembly process is partly guided by selective linkers. The dynamics of the linking process result from a reaction–diffusion process (diffusion, because the linkers diffuse in the two-dimensional membrane bounding the vesicle).
• Data interpretation: A central challenge in bringing modelling and experimentation to true collaboration is bridging the gap between model variables and experimental observables. Given the experimental constraints, we can often access quantities such as the temporal behaviour of some concentrations only indirectly (e.g., by analyzing fluorescence images). We must develop or at least customize techniques to achieve this in the context of Bio-HhOST. Ideally, a simulation in Bio-HhOST is calibrated by experimental data and then reliably delivers experimentally difficult-to-access quantities such as concentration values. Right now, we estimate reaction parameters from experimental data. This is done by invoking some optimization process. The available data results in an underdetermined system and various combinations of parameters reproduce the data. This situation is known in modelling, but presently, no AI–method is known to handle the occurrence of distributions of parameter sets instead of one optimal parameter set, at least in the context of experimental life science.