Feature Selection for Anti-Cancer Plant Recommendation

Every year tremendous experimental analysis has been done for evaluation of anti-cancer properties of plants. A good ranked list of potential anti-cancer plants which raised out of veri ed anti-cancer metabolites, reduces the time and cost for evaluating plants; otherwise, we charged for testing unrelated plants. Ranked list produced by analyzing plant-metabolite biological graphs are candidate for such situation. Graph nodes are ranked according to some graph features. A problem with this approach is how to select the good features of graphs. In this paper a metric used in information retrieval and recommender systems is employed for comparing two di erent ranked list. In an information retrieval system such as search engines, a good system should show the top results rst. A metric named Average Precision is used here for discriminating di erent lists, resulted from di erent features. We build a network of similarity of plants according to their common metabolites. After that, with various combinations of the graph features, the plants are ranked. The subset of features which produces the ranked list with higher AP score is considered as the best features for anti-cancer plant recommendation. The proposed method could be employed to select the best graph features in screening of anti-cancer plants from an unveri ed plants list. So that, the plant with higher score in the list have higher chance to have anti-cancer properties.