Computational Efforts Generation of Hypotheses

One way to develop an in silica tool to predictive promiscuity is to apply a NB classifier for modeling, a technique that compares the frequencies of features between selective and promiscuous sets of compounds. Bayesian classification was applied in many studies and was recently compared to other machine-learning techniques [26, 27, 43, 51, 52]. 

In order to classify promiscuous and selective compounds, we used the NB modeling protocol available in Pipeline Pilot (Scitegic) [53]. The data was split randomly into 5193 compounds for modeling and 574 compounds for testing the models. In addition to the test set, 302 known drugs were also profiled and kept separate for testing the models. All sets were checked visually to ensure that no chemical classes were overrepresented in one set or the other. 

The specificity and sensitivity of each model is reported in the middle and bottom ofTable 13.1. In general, the models trained on only Scitegic fingerprints (Model FP) perform better than the other models. The combined score from Model FP and Model PG improve the prediction on the drug set which contains molecules chemically different that the test set and the training set. A relatively high enrichment was observed for both models although the selectivity model appears more accurate than the promiscuity model. 

Combined score = (score from Model PC 10) + score from Model FP 

To validate the classification models, the sensitivity SE and specificity SP of an individual model were evaluated by the equations  

---