GIST Repository
College of Information and Computing
Department of Electrical Engineering and Computer Science
4. Theses(Ph.D)
Development of prediction models for risk of anti-cancer drug
- Abstract
- In modern medicine, personalized cancer treatment aims to balance drug efficacy with patient safety. Accurately predicting drug responses in cancer cells while minimizing adverse side effects is critical for optimizing anticancer therapies. This dissertation addresses this challenge through two key studies. The first study focuses on developing a predictive model to estimate drug responses in cancer cells. This model leverages advanced machine-learning techniques to analyze large-scale genomic data, enabling accurate predictions of which drugs may be most effective for specific cancer types. The study emphasizes the importance of identifying existing drugs that can be repurposed for alternative cancer treatments, thereby facilitating the drug discovery process. This approach has the potential to reduce the time and cost associated with developing new drugs while improving the success rate of treatment outcomes for cancer patients. However, selecting drugs based exclusively on their effectiveness against cancer cells may overlook the potential for adverse side effects. To address this, the second study introduces a transformer-based model, Crossfeat, which predicts both the occurrence and frequency of drug side effects. Unlike existing models that rely on historical data, Crossfeat independently predicts side effect frequencies, making it more applicable to new drugs. By estimating the likelihood of side effects, this model supports the identification of drugs that are both effective against cancer and carry reduced risks of harm to patients. Together, these models form a comprehensive framework for selecting anticancer therapies that are not only effective but also minimize side effects. The results of these studies have the potential to significantly enhance personalized medicine by offering more precise and safer cancer treatments. This research represents a step forward in advancing AI-driven methods in pharmacogenomics, with broad applications in both clinical practice and pharmaceutical development.
- Issued Date
- 2025
- Type
- Thesis
- Alternative Author(s)
- Bin Baek
- Department
- 대학원 전기전자컴퓨터공학부
- Advisor
- Lee, Hyunju
- Table Of Contents
- Abstract i
List of Contents iii
List of Tables v
List of Figures vi
1 Introduction 1
1.1 Current status and importance 1
1.2 Study objectives and research goals 2
1.3 Significance and potential impact of the study 4
2 Related works 5
2.1 Comprehensive approaches in multi-omics integration for drug response prediction 5
2.2 Modeling the occurrence of drug side effects 6
2.3 Dependence of drug side effect frequency prediction on historical data 7
3 Integrated drug response prediction models pinpoint repurposed drugs with effectiveness against rhabdomyosarcoma 10
3.1 Introduction 10
3.2 Methods 11
3.2.1 Dataset and pre-processing 11
3.2.2 Sarcoma cell lines 13
3.2.3 Approaches for drug response prediction 15
3.3 Results 19
3.3.1 Prediction performance of the AE-NN model for GDSC cell lines 19
3.3.2 Drug response prediction and drug selection for RMS cell lines 19
4 Crossfeat: a transformer-based cross-feature learning model for predicting drug side effect frequency 24
4.1 Introduction 24
4.2 Methods 25
4.2.1 Benchmark dataset 25
4.2.2 Construction of input features 26
4.2.3 A transformer-based cross-feature learning model for drug side
effect frequency prediction (CrossFeat) 28
4.2.4 Experimental design 34
4.2.5 Independent FAERS SI dataset 36
4.3 Results 40
4.3.1 Comparison of model performance 40
4.3.2 Predictive performance across drug side effect frequencies 42
4.3.3 Ablation study 44
4.3.4 Variation of transformer encoder 47
4.3.5 Case studies 49
4.3.6 Evaluation with independent dataset 50
5 Conclusion 55
References 57
Supplementary Information 62
- Degree
- Doctor
- Appears in Collections:
- Department of Electrical Engineering and Computer Science > 4. Theses(Ph.D)
- 파일 목록
-
-
관련 파일이 존재하지 않습니다.
-
Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.