Machine learning (ML) and in particular AI generative methods are transforming drug discovery and development. One of their key limitations is related to training set data narrowing the applicability domain of the approach and its ability to extrapolate to novel research areas where data may be limited. On the opposite physics-based methods based on first principles molecular description of the system provide a rigorous solution complementary to ML-based solutions. They can provide an effective strategy to create synthetic data to extend the ML approach training set or to validate the hypothesis and results proposed by an AI strategy. Examples of the impact of combining physics-based solutions, like absolute free energy methods, with active learning and knowledge graph strategies will be presented.