Theory Article Volume 6, Issue 9 pp 731—741

Figure 2. Biological activity spectra of the gerosuppressant olive oil oleuropein OA. The results of predicted activity spectra generated by PASS are presented as a bar graph of biological activities with the probabilities “to be active” (Pa) and “to be inactive” (Pi) calculated for each activity. The values vary from 0.000 to 1.000; the higher a Pa value is the lower is the predicted probability of obtaining false positives in biological testing. The lists are arranged in descending order of Pa-Pi; therefore, more probable biological activities are at the top of the list. The list can be shortened at any desirable cutoff value, but PASS uses the criteria Pa=Pi as the as the default threshold, i.e., only biological activities with Pa > Pi are considered as probable for a particular compound. If we choose to use rather high value of Pa as cutoff for selection of probable activities, the chance to confirm the predicted activities is high too, but many existing activities will be lost. For instance, if one selects for consideration particular biological activities predicted with Pa > 0.9, then about 90% of actual activities will be lost (i.e., the expected probability to find inactive compounds in the selected set is very low but about 90% of active compounds will be missed). If one lowers the Pa threshold to 0.8, the probability to find inactive compounds is still low, but about 80% of active compounds will be missed, etc. Another important aspect of PASS predictions is the compounds' novelty. If one limits to high Pa values, one may find close analogues of known biologically active substances among the tested compounds. For instance, for Pa > 0.7, the chance to experimentally find the biological activity is high, but some of the activities may be close analogue of known pharmaceutical agents. If one chooses 0.5<Pa<0.7 values, the chances of obtaining activity in the experiment are lower, but the compound may be less similar to known pharmaceutical agents. For Pa < 0.5, the chances of obtaining activity in the experiment are even lower, but if activity is found, the compound might happen to be a new chemical entity. Nevertheless, it is important to keep in mind that the probability Pa reflects the similarity of a molecule under prediction with the structures of molecules, which are the most typical in a sub-set of “actives” in the training set. Therefore, there is no usually direct correlation between the Pa vales and quantitative characteristics of biological activities.