الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 131 |  |  |
| | | from | 131 | | |
Abstract
Discovering possible Drug Target Interactions (DTIs) is a decisive step in the
detection of the effects of drugs as well as drug repositioning. There is a strong incentive
to develop effective computational methods to predict potential DTIs, as traditional DTI
laboratory experiments are expensive, time-consuming, and labor-intensive. This thesis
aims to predict drug interactions in human cells with the most accurate performance
using machine learning techniques.
First, DTIs prediction model is developed to take advantage of the structured form
of proteins and drugs. This model obtains features from protein amino-acid sequences,
and from drugs SMILES (Simplified Molecular Input Line Entry System) strings using
encoding techniques. Empirical results show that our model based on ensemble learning
algorithms for DTI prediction provide more accurate results from both structures and
features data. Second scheme predicting DTIs based on drug chemical structures and
protein sequences are employ to extract the drugs and protein’s characteristics. Then,
proposed approach discovering negative samples using a support vector machine one-
class classifier to tackle the imbalanced data problem. Negative and positive samplings
were constructed and fed into different prediction algorithms to identify DTls. Third,
Network-based approach is developed utilizing feature representation and deep learning.
This approach extracts the relevant features of drugs and proteins from heterogeneous
networks using graph mining techniques, by constructing a heterogeneous graph of
known drug-protein interactions, protein-protein, and drug-drug similarities. Then
applying two feature extraction techniques, the first utilizes the cosine similarity
coefficient and random walk with restart model, and the second involves a similarity
selection procedure and a similarity fusion algorithm. Four Benchmark datasets are used
to evaluate the proposed approach.
Finally, the hierarchical network embedding methods, and the graph autoencoder
(GAE) scheme developed to extract the embedding feature vectors of drugs and
numbered targets from multi-source heterogeneous networks to predict DTIs. In
comparison with literature methods, the achieved results demonstrate the validity and
effectiveness of the proposed work for drug target interactions and improvement of the
drug repositions