The r_m ² metrics for validation of QSAR/QSPR models

Quantitative structure-activity/property relationship (QSAR/QSPR) models being used for the prediction of activity/property of untested chemicals can be exploited for the prioritization plan of synthesis and experimental testing. Validation of QSAR models plays a crucial role for the selection of robust and best predictive models that can be utilized for future activity/property prediction of new molecules. There exists a number of metrics to express the performance of a model but traditionally QSAR models are validated based on classical metrics for internal (Q²) and external validation (R² _pred). But being primarily dependent on the mean activity data of the training set compounds, both the metrics tend to achieve acceptable values (>0.5) whenever a data set with a wide range of activity/property data is considered. However, such values may not truly reflect the quality of predictions. Therein lies the utility of the r_m ² metrics developed by Roy et al., which consider the actual difference between the observed and predicted response data without consideration of training set mean, thereby serving as a more stringent measure for the assessment of model predictivity compared to the traditional validation metrics. The r_m ² metrics depend chiefly on the difference between the observed and predicted activity data and convey more accurate information regarding the quality of predictions. Thus, the r_m ² metrics strictly judge the ability of a QSAR model to predict the activity/property of untested molecules. We herein focus to have an overview of evolution of different r_m ² metrics and the software tools for their calculation followed by their successful applications in QSAR modeling.