Rank-ordered biological data

According to the nature of the activity data, QSAR studies can be divided into continuous (activities, i.e., response variable, takes many different values from within some interval), category (activities are represented by ranks or ordinal numbers), and classification (activities are different types of biological properties which cannot be rank ordered) approaches. 

Each active molecule in a data-set was taken in turn and its similarity calculated with each of the other molecules in that data-set. The molecules were then ranked in decreasing order of similarity and cut-offs applied to retrieve the top 5, top 10 and top 20 compounds from the ranking. These were then checked to determine whether they were active or inactive in the particular biological test system associated with that data-set. The overall utility of a similarity measure was calculated by taking the mean numbers of active molecules at the three cut-off positions when averaged over all of the active compoimds. 

---