Retrospective virtual screening

Xu H, Agrafiotis DK Retrospect and prospect of virtual screening in drug discovery. Curr Top Med Chem 2002 2 1305-20. 

One of the goals of QSAR studies is to help explain retrospectively the response or property of a molecule with a rationale based on molecular structure. A second major goal and challenge of QSAR or QSPR studies is to develop models that are able to predict quantitatively the property of new molecules either real or virtual compounds. Thus, successful predictive QSAR models can have a tremendous impact in the design of new molecules. Furthermore, predictive models are useful to perform in silico predictions of the properties of new structures. In virtual screening, those molecules that are predicted to have the desired property according to the QSAR model are selected as best candidates. Reviews, examples, caveats, and modified versions of QSAR are described elsewhere (Kubinyi, 1997a,b Wermuth, 2008). Some recent examples reported in the food chemistry field are summarized in Table 2.4. 

If the virtual screening process is successful, the bioactive compounds will be enriched in the upper part of this ranked list. Enriched means that there are more active compounds in this part of the list than that one would expect based on the average content of bioactive compounds in the searched database. This enrichment can be determined only in retrospective studies where both the active and the inactive compounds in the database are known The emichment factor is typically used in such studies as a quality criterion for a particular virtual screening approach. Since the emichment factor depends on the chosen subset of the ranked list, it is common to calculate enrichment curves where the cumulated number of hits is plotted against the ranked list of compounds in the search database. In recent years, more and more researchers prefer other types of enrichment plots (e.g., ROC curves). The interested... 

The simplest way to compare the performance of different descriptors is to perform a retrospective study and to count the numbers of true actives found in different virtual screening hit lists of predefined size and compare them with the hit rate in the search database (enrichment). Enrichment factors, however, are not sufficient to assess the scaffold hopping potential of a descriptor or a search method. 

Retrospective Validation Prior to Prospective Virtual Screening... 

Although an increasing number of prospective virtual screening outcomes are now reported, most studies are still retrospective with a focus on novel software or virtual... 

The feasibility of GPCR homology models for structure-based virtual screening was initially demonstrated in a few retrospective [65,66] and prospective applications [67, 68[. Varady et al. [69] reported on the discovery of novel and potent dopamine D3 ligands from a hybrid pharmacophore- and structure-based approach. Of the 20 tested compounds, 8 showed K values <1 [xM. 

Only prospective virtual screenings are shown purely retrospective studies or enrichment studies, as well as studies without experimental testing, are not included. The targets are grouped in different protein classes and sorted alphabetically within each class (first column). Where possible, the number of tested compounds, the number of detected active compounds, and activity data of the most potent hits are reported in the third column (n.a., not available). In general, the results refer to the primary screening hits secondary screens and optimizations are not considered. 

Retrospective experiments are commonly used to test the ability of virtual screening techniques to retrieve known bioactive molecules from a background of assumed inactive molecules. Suchabilityisstronglytargetandtechniquedependentandthusone should in principle look for validation results that are relevant for the studied target. 

---