In experimental Life Sciences, simulations are needed in order to extrapolate costly experiments and to design decision-support tools. When extensive mathematical knowledge is not available at the desired scale, expertise and data-driven models can be used as a basis for these simulations. Decision tree algorithms are efficient approaches for data-driven discovery of complex non obvious relationships. Their readability and the absence of a priori assumptions make them particularly useful for variable selection in highly multidimensional problems, therefore they are ideal to display statistically important variables on which the expert should focus. However, to be really useful, such models must be trusted by their users. From this perspective, the domain expert knowledge can be collected and modelled to help guiding the learning process and to increase the confidence in the resulting models, as well as their relevance. We propose a generic iterative approach to design ontology-aware and relevant data-driven models. It is based upon an ontology to model the domain knowledge, and it uses decision trees as the learning method. Relationships between concepts from the ontology and variables from the data sets are formalized and various data processing techniques are presented to build more significant variables from the original ones, exploiting both the ontology and expert feedback. Subjective and objective evaluations are both involved in the process. Starting from an initial data set, an initial data-driven model is learnt (step 1). This model is first evaluated with numerical criteria, and submitted to domain experts, who may enrich the ontology by suggesting new relations between some variables (step 2). Data transformations are applied according to these new relations (step 3). The whole process is repeated iteratively. A case study concerning the impact of agri-food transformation processes on the nutritional quality of wheat-based products is presented to demonstrate the interest of this approach.