The preparation of hit-rich virtual libraries can be a time-consuming and challenging process. Incorrect characterization of drug-like properties will result in low HTS payoffs, with fewer, less promising hits. While scores of 2D and 3D descriptors can be rapidly calculated or predicted, these properties are of limited use unless they are correlated with drug action. Furthermore, in order to protect against over-fitting, any predictive relationship must use only a reasonable subset of descriptors with minimal covariance.
Quantitative structure-activity relationships (QSAR) and statistical modeling can be used to develop such predictive relationships, greatly improving the profile of virtual libraries. Chemically aware statistical modeling software combines sophisticated analysis tools with the ability to easily visualize the structures and properties used to derive a structure-activity relationship. QSAR is one of the most mature techniques in rational drug design, and has repeatedly proven itself to be a low-cost, high-return investment.