Chapter 9: Assessing Quantitative Model Quality and Performance Check Access

Quantitative, predictive methods applied to drug design aim to provide numerical estimates of compound properties for compounds yet to be synthesised. Both physics-based and empirical quantitative structure–activity relationships (QSARs) have been used. Empirical quantitative structure–activity relationships have been widely adopted due to their ease of generation and their successful application to drug design over many years. But ease of generation of QSAR models is confounded by the complexity of defining the confidence with which we can extrapolate predictions. Much of the science of QSAR is predicated on an attempt to assess the quality of predictions and the extent to which the model can predict beyond the model's own domain of applicability. This chapter will describe common diagnostic statistics of model fit, predictivity, model robustness and model confidence, how they are used, and sometimes abused, and current state-of-the art in defining model quality and performance.