In this paper, we describe a formal logical framework which we claim as essential to prove and to revise a model produced by combined ILP techniques. The dynamic process of proof embrace the supervision of the learning machine by a human, and this framework places the interpretation of contradictions in the heart of the interactive process which leads to a model which can be discussed, justified, and proven. We illustrate and validate this framework on an industrial application in the field of Drug Discovery, combining different learning processes to predict pharmaco-kinetic properties (ADME-T) and adverse side effects of therapeutic drug molecules.