How complex does SAR have to be in order to explain data? How many parameters are required? Are hundreds of parameters (descriptors) justified? In some cases, even simple nearest neighbors (NN) can give suggestive results. SAR is based on similarity and similar molecules often have similar properties. What happens if NN models are based on all neighbors (global) and use a non-linear weighted average?  What happens if we build these regression models with very few free parameters?  Here we investigate just such an approach and find that simple kernel regression of molecular similarities can provide predictions comparable to more complex models with hundreds of adjustable parameters.

One parameter QSAR: Prediction by general similarities (PGS)

248th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.