Comparison of 2D fingerprint types and hierarchy level selection methods for structural grouping using Ward's clustering

Four different two-dimensional fingerprint types (MACCS, Unity, BCI, and Da ylight) and nine methods of selecting optimal cluster levels from the outpu t of a hierarchical clustering algorithm were evaluated for their ability t o select clusters that represent chemical series present in some typical ex amples of chemical compound data sets. The methods were evaluated using a W ard's clustering algorithm on subsets of the publicly available National Ca ncer Institute HIV data set, as well as with compounds from our corporate d ata set. We make a number of observations and recommendations about the cho ice of fingerprint type and cluster level selection methods for use in this type of clustering.