engleski

Significance verification and simplification of some often used structure-property models in molecular biosciences

In development of structure-property/activity models, very often, too many parameters have been introduced, i.e. more than it could be justified by performing significance testing. Such a practice could be initiated by an intention of authors to improve the model accuracy over the previously results, published either for similar, or for the same problem or data set, because it is desirable to have (numerically) better results than the best previous ones. We will illustrate how structure-property/activity models can be simplified and improved by performing significance testing of optimized parameters introduced in final models, on two examples (1) modeling based on connectivity indices, and (2) modeling folding/unfolding rates of proteins.

-.

nije evidentirano