Application to Library Design

A straightforward method for PCA is the NIPALS (nonlinear iterative partial least squares) algorithm it is quickly implemented and can be applied to large datasets [58, 79], 

Now the whole trick of this procedure is the following. Prior to PCA, artificial compounds (or a real set of reference compounds) are added to the data matrix X. These additional data points (library spikes ) have coordinates that represent idealized properties of the library. For example, if the aim is to generate a library for the cannabinoid receptor family, possible coordinates of the idealized artificial compound might be 1 for drug-likeness, 1 for GPCR-ligand-likeness, 1 for cannabinoid-likeness, 0 for dopamine-likeness, 0 for kinase-inhibitor-likeness, and so forth. In this example, the value T indicates the maximum value of a property (presence of a feature), 0 indicates minimum values (absence of a feature). Of course, appropriate prediction models must be at hand. 

PCA is not the only projection method that can be used. Various types of nonlinear projections have been employed, e.g., Sammon mapping and nonlinear PCA [80], and several software packages can be used to graphically visualize library distributions and aid compound selection [81]. 

Fast binary filtering methods can also be used for scaffold ranking, i.e., the prioritization of combinatorial scaffolds based on predicted properties. Privileged scaffolds were selected to demonstrate this idea [82], Piperazines SI, benzodiazepines S2, and spiroindolines S3 have been described as GPCR-privileged scaffolds [83], Scaffold S4 represents a SPIKET motif for tubulin binding which is effective for inhibiting cellular proliferation [84], Dysidiolide-derived compounds 

---