Predicting drug synergy using data mining

Official URL: http://risc01.sabanciuniv.edu/record=b1640581 (Table of Contents)

Antibiotic resistance has become an important health threat across the world during recent years. One of the solutions to reduce antibiotic resistance is to nd ways in order to use e cient amounts of antibiotics in treatments. It has been seen that some antibiotics are synergistic, i.e, if they are administered together, they will boost the individual antibacterial and antifungal e ects. Identi cation of synergistic antibiotics can be of signi cant assistance to medical practitioners in order to optimize the amount of antibiotics to be used. In this thesis we have conducted a set of analyses using data mining based approaches. Chemogenomic pro les and chemical properties of drugs have been utilized to predict synergy between them. Two datasets, E. Coli and yeast were used in order to perform the analysis. GRASP meta-heuristic algorithm was implemented on chemogenomic features in order to predict synregies which yielded in 0.94 accuracy and 0.82 Area Under ROC curve for E. Coli dataset. In order to further explore the chemogenomic features, we suggest a novel algorithm to predict synergy. This algorithm resulted in Area Under ROC curve and accuracy of 0.71 and 0.91, respectively for E. Coli dataset. Next, two chemical features, XLogP3 and Q PC- were used to perform the analysis by employing decision trees and random forest classi ers. Our analysis indicate that Q PC- chemical feature can be as discriminative as XLogP3 which has been used in literature previously. Employing chemical features resulted in most accurate prediction among the implemented methods. In this thesis, details of the above-stated methods and algorithms will be presented.