Interpretability from a new lens: Integrating Stratification and Domain knowledge for Biomedical Applications