Enhance explainability of manifold learning.