Computational approaches to study drug resistance mechanisms

Official URL: http://risc01.sabanciuniv.edu/record=b1655304 (Table of Contents)

Drug resistance is a major obstacle faced by therapists in treating complex diseases like cancer, epilepsy, arthritis and HIV infected patients. The reason behind these phenomena is either protein mutation or the changes in gene expression level that induces resistance to drug treatments. These mutations affect the drug binding activity, hence resulting in failure of treatment. All this information has been stored in PubMed directories as text data. Extracting useful knowledge from an unstructured textual data is a challenging task for biologists, since biomedical literature is growing exponentially on a daily basis. Building an automated method for such tasks is gaining much attention among researchers. In this thesis we have developed a disease categorized database ZK DrugResist that automatically extracts mutations and expression changes associated with drug resistance from PubMed. This tool also includes semantic relations extracted from biomedical text covering drug resistance and established a server including both of these features. Our system was tested for three relations, Resistance (R), Intermediate (I) and Susceptible (S) by applying hybrid feature set. From the last few decades the focus has changed to hybrid approaches as it provides better results. In our case this approach combines rule-based methods with machine learning techniques. The results showed 97.7% accuracy with 96% precision, recall and F-measure. The results have outperformed the previously existing relation extraction systems thus facilitating computational analysis of drug resistance against complex diseases and further can be implemented on other areas of biomedicine. Literature is filled with HIV drug resistance providing the worth of training data as compared to other diseases, hence we developed a computational method to predict HIV resistance. For this we combined both sequence and structural features and applied SVM and Random Forests classifiers. The model was tested on the mutants of HIV-1 protease and reverse transcriptase.Taken together the features we have used in our method, total contact energies among multiple mutations have a strong impact in predicting resistance as they are crucial in understanding the interactions of HIV mutants. The combination of sequence-structure features o↵ers high accuracy with support vector machines as compared to Random Forests classifier. Both single and acquisition of multiple mutations are important in predicting HIV resistance to certain drug treatments. We have discovered the practicality of these features; hence these can be used in the future to predict resistance for other complex diseases. Another way to deal drug resistance is the application of drug repurposing. Drug often binds to more that one targets defined as polypharmacology which can be applied to drug repositioning also referred as therapeutic switching. The traditional drug discovery and development is a high-priced and tedious process, thus making drug repurposing a popular alternate strategy. We have proposed a method based on similarity scheme that predicts both approved and novel targets for drug and new disease associations. We combined PPI, biological pathways, binding site structural similarities and disease-disease similarity measures. We used sixty drugs for training the algorithm and tested it on eight separate drugs. The results showed 95% accuracy in predicting the approved and novel targets surpassing the existing methods. All these parameters help in elucidating the unknown associations between drug and diseases for finding the new uses for old drugs. Hence repurposing offers novel candidates from existing pool of drugs providing a ray of hope in combating drug resistance.