Browsing by Author "SULTAN, AHMED ATTA AHMED"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Machine Learning Techniques for Pharmaceutical Bioinformatics
    (The British University in Dubai (BUiD), 2018-11) SULTAN, AHMED ATTA AHMED
    This dissertation presents a novel drug classifier to automate the prediction of drug indication and drug interactions with other drugs. The study integrates knowledge visualization, analysis, as well as development of a predictive model based on the Drug-Drug Interactions (DDIs) as a complex network. DDIs network analysis reveals unique drug features and explains unknown drug behaviors. Each drug molecule has a unique chemical structure and a set of pharmacological features. This set of attributes imposes how each drug performs its action inside a human body. Drug molecule interacts with multiple components in the biological system, for example, enzymes, proteins, among other drugs. The complexity of the chemical and pharmacological features forces the interaction between drug molecule and all other entities in the biological system to follow specific rules. The full features for each drug are not fully explained by researchers due to the incomplete drug profile description. DDIs network has a significant role in drug repurposing; it uncovers the hidden properties of the drug behavior. Predicting drug properties is presented as a contribution effort to drug repositioning approach. To confirm the visual analysis, a binary matrix is drawn from each drug profile based on DDIs dataset. In this matrix, each drug is represented by a vector of attributes from all other drugs. A predictive model is developed to predict drug indication as well as to predict new DDIs using multiple machine learning algorithms. This dissertation presents a case study of predicted anti-cancer activity for 38 drugs. The proposed Artificial Intelligence approach for drug-related properties prediction demonstrates a high potential in complementing the current computational techniques. The predicted anti-cancer activity is computationally validated by a 10-fold cross validation evaluation technique and clinically supported by extensive literature review confirming the achieved results. In conclusion, the predicted drug features can provide new directions towards promising candidates for drug repositioning.