Effect of Parameter Variations on the Effectiveness of HQSAR Analyses
Michael Seel, David B. Turner, Peter Willett · Quantitative Structure-Activity Relationships · 1999
HQSAR is a new method for generating alignment-free quantitative structure-activity relationships. Experiments with four different datasets suggest that the variations in PLS scores that are observed with short hologram lengths can be substantially removed either by taking the mean or median of the crossvalidation scores or by using very long holograms. However, because the hashing process unnecessarily obfuscates PLS regression modelling, we suggest that PLS should preferentially be applied to unhashed fragment bit-strings where computationally feasible. The predictive ability of the method is also affected by the size of the fragments that are used, although this effect appears to be dataset-dependent. Variations in the type, rather than the size, of the fragments utilised can also have a significant effect on both internal and external predictivity scores.