Use of the R-group descriptor for alignment-free QSAR

An R-group descriptor characterises the distribution of some atom-based property, such as elemental type or partial atomic charge, at increasing numbers of bonds distant from the point of substitution on a parent ring system. Application of Partial Least Squares (PLS) to datasets for which bioactivity data and R-group descriptor information are available is shown to provide an effective way of generating QSAR models with a high level of predictive ability. The resulting models are competitive with the models produced by established QSAR approaches, are readily interpretable in structural terms, and are shown to be of value in the optimisation of a lead series.