Deep Learning Approaches Towards Computerized Drug Discovery Público

Abstract

Proposing a new drug candidate is an essential part of the drug discovery process, consisting of many sub-tasks. Traditionally, these tasks have been tackled by chemistry and pharmaceutical experts and take years to design. Therefore, this thesis aims to accelerate drug discovery by proposing deep-learning models that accomplishes these tasks effectively and quickly. For the target identification problem, we propose new feature selection methods for both disease-related and prognosis-related features. Next, we propose a new drug-target interaction model to perform the drug re-purposing task. In this model, we present a new molecule representation to overcome the limitation of the current models. We also propose a novel drug candidate generation model that can modify an existing drug to meet given molecule properties. For each project, we present an empirical evaluation to show the competency of the proposed approaches. In addition, we also provide analyses or case studies to demonstrate the practicality of our approaches. 

Table of Contents

1 Introduction 1

1.1 Drug Discovery Pipeline ......................... 2

1.2 AI in Drug Discovery........................... 4

1.3 Contributions ............................... 5

1.4 Outline................................... 7

2 Disease-Related Target Identification 9

2.1 Motivation................................. 9

2.2 Problem Definition ............................ 12

2.3 Proposed Model:Wx........................... 13

2.4 Experiments................................ 17

2.5 Discussion................................. 23

2.6 Contribution................................ 26

3 Prognosis-Related Target Identification 27

3.1 Motivation................................. 27 

3.2 Problem Definition ............................ 30 

3.3 Proposed Model:CascadedWx ..................... 31 

3.4 Experiments................................ 33 

3.5 Discussion................................. 35 

3.6 Contribution................................ 36

4 Drug Repurposing 37

4.1 Introduction................................ 37

4.2 Related Work ............................... 41

4.3 Problem Definition ............................ 42

4.4 Proposed Model: Molecule Transformer-Drug Target Interaction . . . 42

4.4.1 ModelArchitecture........................ 43

4.4.2 Molecule Transformers ...................... 44

4.4.3 Protein CNNs........................... 49

4.4.4 Interaction Denses ........................ 49

4.5 Experiments................................ 50

4.5.1 Datasets.............................. 50

4.5.2 Training Details.......................... 51

4.5.3 Evaluation Metrics ........................ 53

4.5.4 Baselines.............................. 54

4.5.5 Results............................... 55

4.6 Case Studies................................ 56 

4.6.1 Anticancer Drug Discovery.................... 56 

4.6.2 Antiviral Drug Discovery..................... 58

4.7 Discussion................................. 63

4.8 Contribution................................ 64

5 Molecule Generation 65

5.1 Introduction................................ 65

5.2 Related Work ............................... 68

5.3 Problem Definition ............................ 70

5.4 Proposed Model: Controlled Molecule Generator . . . . . . . . . . . . 70

5.4.1 Model Overview.......................... 71 

5.4.2 Background ............................ 71

5.4.3 Molecule Translation Network.................. 73

5.4.4 Constraint Networks ....................... 75

5.4.5 Modified Beam Search with Constraint Networks . . . . . . . 79

5.4.6 Diversifying the Output ..................... 81

5.5 Experiments................................ 81

5.5.1 Datasets.............................. 82

5.5.2 Pre-Training of Constraint Networks . . . . . . . . . . . . . . 84

5.5.3 Single Objective Optimization.................. 84

5.5.4 Multi Objective Optimization .................. 87

5.5.5 Ablation Study .......................... 90

5.6 Case Study ................................ 91

5.7 Discussion................................. 92

5.8 Contribution................................ 93

6 Conclusion 94

7 Future Direction 96

Bibliography 97

About this Dissertation

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Laney Graduate School
Department	Computer Science and Informatics
Degree	Ph.D.
Submission	Dissertation
Language	English
Research Field	Computer Science Biology, Bioinformatics
Palavra-chave	drug discovery molecule biomarker deep learning protein
Committee Chair / Thesis Advisor	Ho, Joyce C., Emory University
Committee Members	Xiong, Li, Emory University Sun, Jimeng, University of Illinois Qin, Zhaohui S., Emory University

Última modificação

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Deep Learning Approaches Towards Computerized Drug Discovery ()	2020-04-07 14:50:58 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions