With more than 9000 scientific articles publishedeach day, the scientific literature, includingexperimental data and knowledge, is a valuable sourceof information for developing predictivemodels to design decision support systems. However,scientific data are heterogeneously structured,found mainly in text format and expressed using differentvocabularies. This study developed apractical and versatile pipeline that combines ontology,databases and computer calculation toolsbased on the theory of belief functions and Bayesiannetworks. The objective was to design aniterative process that can structure a domain of knowledge,use domain ontology to integrate data instructured databases, and build a predictive modelfrom ontology and data. The ontology paradigm isused to help integrate data from heterogeneous sourcesand build the structure of the Bayesian network. The parameters of the Bayesian network are estimated using the integrated data taking intoaccount their reliability. The pipeline can be usediteratively to enrich the database and the modelwithnew data and knowledge over time without damagingthe entire system. The pipeline was assessedby applying it to a complex food engineering process:skimmed milk microfiltration. , which is one ofthe first operations that can be carried out afterharvesting. This application enabled buildingthedomain ontology of milk microfiltration to annotatestate-of-the-art literature sources, creatingastructured database and developing a predictivemodel of the permeation flux associated withoperating conditions.