Correlation analysis between watersolubility, octanol-water partition coefficient and melting point based on clustering

By means of clustering according to structure similarity one can manage easily large databases containing several Lens of thousands of chemicals. A new methodology in which statistical criteria are used to decide the optimal threshold in the clustering process is proposed. Clustering defined the several chemical classes that were present in our training set. All the clusters except one showed a correlation of logWS with logK(ow) and melting point. The latter covers all chemicals with melting points below room temperature, resulting in a logWS-logK(ow) relationship. This cluster showed also a weak correlation. probably due to the insufficient number of available screens. Such a limited number of screens permits that rather different chemicals share the same cluster. Using statistical criteria, our approach resulted in 3 QSARs with reasonably good predictive capabilities, The models resulting from the smaller clusters are characterised by high correlation coefficients, describing the cluster itself very well. Due to our stringent bootstrap criterion, they perform close to random models as indicated by the statistical characteristics. Some clusters showed rather low correlations. The well behaved models proved their usefulness through external validation. The logWS-values calculated using our QSARs agreed within 1 log-unit with these reported in the literature.