Gaussian process: an efficient technique to solve quantitative structure-property relationship problems.

We introduce the Gaussian process (GP) model for the empirical modelling of the log P values of 44 1,2-dithiole-3-one molecules. A brief theoretical description of the method is given. Descriptive and predictive abilities of GP are evaluated and compared to multilinear regression results. Special attention is devoted to the automatic relevance determination (ARD) to reduce input variable numbers, which avoid the use of principal component analysis. The present approach was found to be an efficient method and a good alternative to more complicated using artificial neural network systems.