Outlook Unsupervised learning and diversity considerations

We did not yet discuss the choice of the real Ubrary. In the examples treated above structures and properties were taken from databases or from the literature. Ideally, the choice of a real library should be a basis for successful optimization of an experiment in combinatorial chemistry. The real libreuy then should contain structures of as high diversity as possible. 

In our applications in combinatorial chemistry, independent variables are molecular descriptors, and their values may be taken either from the building blocks or from the compounds in the libreuy themselves. Important methods of non-supervised learning are principal component analysis (PCA) euid cluster analysis. 

The main objective of QSPRs and QSARs, as covered in this chapter, is to start from the molecular structure and arrive at a predicted property value. However, the opposite also plays an important role in chemoinformatics, namely starting from a given property (or properties) and trying to predict the molecular structure. The next chap- 

---