Virtual screening (VS) is a computational technique used in the drug discovery to identify the binding strength of potential drug molecules to given targets. In the structure-based VS, candidate molecules are docked to a protein target followed by using a scoring function (SF) to identify binding decoys and true binders. In this talk, I will discuss how mathematical methods, including manifold learning, achieve the top performance in D3R Grand Challenge 3, a worldwide competition in computer-aided drug design and discovery.