語系:
繁體中文
English
說明(常見問題)
回圖書館首頁
手機版館藏查詢
登入
回首頁
到查詢結果
[ subject:"Oncology." ]
切換:
標籤
|
MARC模式
|
ISBD
Advancing Precision Oncology with Em...
~
Choy, Chi Tung.
FindBook
Google Book
Amazon
博客來
Advancing Precision Oncology with Embedding and Deep Learning - From Chemoresistance Forecast to Therapeutic Regimen Proposal.
紀錄類型:
書目-電子資源 : Monograph/item
正題名/作者:
Advancing Precision Oncology with Embedding and Deep Learning - From Chemoresistance Forecast to Therapeutic Regimen Proposal./
作者:
Choy, Chi Tung.
出版者:
Ann Arbor : ProQuest Dissertations & Theses, : 2019,
面頁冊數:
254 p.
附註:
Source: Dissertations Abstracts International, Volume: 81-09, Section: B.
Contained By:
Dissertations Abstracts International81-09B.
標題:
Oncology. -
電子資源:
http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=27783952
ISBN:
9781392354216
Advancing Precision Oncology with Embedding and Deep Learning - From Chemoresistance Forecast to Therapeutic Regimen Proposal.
Choy, Chi Tung.
Advancing Precision Oncology with Embedding and Deep Learning - From Chemoresistance Forecast to Therapeutic Regimen Proposal.
- Ann Arbor : ProQuest Dissertations & Theses, 2019 - 254 p.
Source: Dissertations Abstracts International, Volume: 81-09, Section: B.
Thesis (Ph.D.)--The Chinese University of Hong Kong (Hong Kong), 2019.
With the advance of next-generation sequencing, patient's genome and transcriptome can be profiled easier, faster and cheaper. A number of international effort have been made to gather this information from a large cohort of patients and maintain it in accessible databanks. These large amount of data gradually transform oncology practice into data driven approach, from stratify patients with disease types into suggesting therapy with individuals' biology. However, traditional statistical and computational methodologies have suffered from 'curse of dimensionality' that limit its power on intrinsically high-dimensional biological data. Machine learning, a specific subset of artificial intelligence (AI) that trains a machine how to learn, may overcomes these shortcomings and discovers new information from existing data. In light of this, we explored the application of machine learning algorithm together with traditional computational approach in chemoresistance prediction and therapeutic regimen proposal, which anticipate to improve therapeutic outcome.Firstly, stochastic modeling, a computational simulation methodology, was built to model chemoresistance cell dynamics aiming at predicting the time of chemoresistance arise. The model was calibrated to show clinically relevant output. Sensitivity analysis showed the model was more sensitive to pharmacokinetic coefficients, mutagenicity of drug and growth rate of resistant cells. Hence, microfluidic devices were developed in order to estimate the defining factors of the model. We demonstrated that the device could encapsulate cancer cells in hydrogel.Nonetheless, a single biopsy with limited number of cancer cells would hinder the number of drugs to be tested and insufficient information was obtained for new treatment option. Therefore, we examined the use of machine learning in existing database, and attempt to identify biomarkers for chemotherapy. Machine learning, in particular artificial neural networks (ANNs), have been proven to be powerful in pattern recognition with remarkable accuracy. Based on this, we adopted a method of unsupervised ANN, namely embeddings in the form of collaborative filtering, to extract biological relevant information from The Cancer Genome Atlas (TCGA) gene expression dataset. Ground truth relationship, such as cancer types of the input sample and semantic meaning of genes, were showed to retain in the resulting entity matrices. We also demonstrated the biological relevance of the resulting matrices, and identified potential novel biomarkers for immune checkpoint blockade therapy.The framework was extended and incorporated into a deep learning network model to predict the sensitizer of chemotherapeutics based on transcriptomic data instead of single or several biomarkers. The deep learning model was composed of two modules, drug perturbation predictor (DPP) and drug response predictor (DRP). The DPP and DRP were trained using The Library of Integrated Network-Based Cellular Signatures (LINCS) and Cancer Cell Line Encyclopedia (CCLE) dataset respectively based on the Broad L1000 assay. The model discovered traditional chemotherapy drugs were a poor choice as "sensitizer" and for drug combinations, while histone deacetylase inhibitors (HDACi) could sensitize cancer cells to other drugs. Several in silico predicted pairs were tested and validated in vitro using cell viability assay. The deep learning model could be a promising tools to suggest chemotherapy regimen based on individuals' tumor biology.In conclusion, the study investigated the use of stochastic modeling, embedding and deep learning to forecast the time of chemoresistance development, to identify the biological subgroup of chemotherapy responders and to propose efficacious sequential drug regimen. We believe the study could accelerate the clinical application of machine learning to benefit cancer patients.
ISBN: 9781392354216Subjects--Topical Terms:
751006
Oncology.
Advancing Precision Oncology with Embedding and Deep Learning - From Chemoresistance Forecast to Therapeutic Regimen Proposal.
LDR
:08929nmm a2200277 4500
001
2264435
005
20200423113052.5
008
220629s2019 ||||||||||||||||| ||eng d
020
$a
9781392354216
035
$a
(MiAaPQ)AAI27783952
035
$a
AAI27783952
040
$a
MiAaPQ
$c
MiAaPQ
100
1
$a
Choy, Chi Tung.
$3
3541558
245
1 0
$a
Advancing Precision Oncology with Embedding and Deep Learning - From Chemoresistance Forecast to Therapeutic Regimen Proposal.
260
1
$a
Ann Arbor :
$b
ProQuest Dissertations & Theses,
$c
2019
300
$a
254 p.
500
$a
Source: Dissertations Abstracts International, Volume: 81-09, Section: B.
502
$a
Thesis (Ph.D.)--The Chinese University of Hong Kong (Hong Kong), 2019.
520
$a
With the advance of next-generation sequencing, patient's genome and transcriptome can be profiled easier, faster and cheaper. A number of international effort have been made to gather this information from a large cohort of patients and maintain it in accessible databanks. These large amount of data gradually transform oncology practice into data driven approach, from stratify patients with disease types into suggesting therapy with individuals' biology. However, traditional statistical and computational methodologies have suffered from 'curse of dimensionality' that limit its power on intrinsically high-dimensional biological data. Machine learning, a specific subset of artificial intelligence (AI) that trains a machine how to learn, may overcomes these shortcomings and discovers new information from existing data. In light of this, we explored the application of machine learning algorithm together with traditional computational approach in chemoresistance prediction and therapeutic regimen proposal, which anticipate to improve therapeutic outcome.Firstly, stochastic modeling, a computational simulation methodology, was built to model chemoresistance cell dynamics aiming at predicting the time of chemoresistance arise. The model was calibrated to show clinically relevant output. Sensitivity analysis showed the model was more sensitive to pharmacokinetic coefficients, mutagenicity of drug and growth rate of resistant cells. Hence, microfluidic devices were developed in order to estimate the defining factors of the model. We demonstrated that the device could encapsulate cancer cells in hydrogel.Nonetheless, a single biopsy with limited number of cancer cells would hinder the number of drugs to be tested and insufficient information was obtained for new treatment option. Therefore, we examined the use of machine learning in existing database, and attempt to identify biomarkers for chemotherapy. Machine learning, in particular artificial neural networks (ANNs), have been proven to be powerful in pattern recognition with remarkable accuracy. Based on this, we adopted a method of unsupervised ANN, namely embeddings in the form of collaborative filtering, to extract biological relevant information from The Cancer Genome Atlas (TCGA) gene expression dataset. Ground truth relationship, such as cancer types of the input sample and semantic meaning of genes, were showed to retain in the resulting entity matrices. We also demonstrated the biological relevance of the resulting matrices, and identified potential novel biomarkers for immune checkpoint blockade therapy.The framework was extended and incorporated into a deep learning network model to predict the sensitizer of chemotherapeutics based on transcriptomic data instead of single or several biomarkers. The deep learning model was composed of two modules, drug perturbation predictor (DPP) and drug response predictor (DRP). The DPP and DRP were trained using The Library of Integrated Network-Based Cellular Signatures (LINCS) and Cancer Cell Line Encyclopedia (CCLE) dataset respectively based on the Broad L1000 assay. The model discovered traditional chemotherapy drugs were a poor choice as "sensitizer" and for drug combinations, while histone deacetylase inhibitors (HDACi) could sensitize cancer cells to other drugs. Several in silico predicted pairs were tested and validated in vitro using cell viability assay. The deep learning model could be a promising tools to suggest chemotherapy regimen based on individuals' tumor biology.In conclusion, the study investigated the use of stochastic modeling, embedding and deep learning to forecast the time of chemoresistance development, to identify the biological subgroup of chemotherapy responders and to propose efficacious sequential drug regimen. We believe the study could accelerate the clinical application of machine learning to benefit cancer patients.
520
$a
隨著新一代基因測序技術(next generation sequencing)的發展,研究人 員可以更容易、更快速、更便宜地取得癌症患者的基因組和轉錄組數據。因 此,國際間有大型跨地域的合作,希望藉此從大量患者中收集這類生物大數 據供研究之用。這類大數據逐漸將腫瘤學研究轉型以數據為主(data-driven) 的實踐方式,加深了科學家對於癌症的了解,並革新了疾病類型的分類 (molecular subtyping)和患者治療提案的基礎。然而,傳統的統計方法被受 "維數災難"(curse of dimensionality)的困擾,這種問題限制了其分析這些 高維度生物數據的能力和效果。然而,機器學習(machine learning)這個人 工智能(artificial intelligence/AI)的分支正正可以克服這些缺點並從現有數據 中提取新信息。有鑑於此,我們探索了傳統計算方法與機器學習演算法在化 學治療耐藥性(chemoresistance)預測和化療方案提案中的應用,期望從而提 高化療治療成效,並促成和加快精準癌症治療的臨床應用。首先,我們利用了隨機仿真數學模型(stochastic modeling)以模擬各種 癌細胞在化療藥物影響下的相互作用,期望能夠從中預測化療耐藥性出現的 時間。經校準數模參數(parameter calibration)後,該模型能輸出與臨床觀察 相關的數值。另外,敏感度分析(sensitivity analysis)顯示該數模對藥物代謝 動力學系數、藥物的致突變性和耐藥癌細胞生長率比較敏感。換言之,上述 參數是決定化療耐藥性出現時間的關鍵因素。由於數模普遍而言對於輸入的 參數相對敏感,因而應用了微流體裝置(microfluidics device)以估計患者中 癌細胞在數模系數間存在的差異,從而得出更準確的預測。該裝置成功將癌 細胞包裹在水凝膠(hydrogel)中,而在水凝膠中的癌細胞亦可被放置在多孔 微流體裝置上繼續培養以進行藥物檢測。除此之外,我們研究了機器學習應用於高維度生物數據的可行性,並嘗 試利用機器學習從生物大數據中識別出化療的生物標誌(biomarker)。基於人 工神經網絡(artificial neural network)在模式識別中具有非常高的準確性,所 以我們採用了一種無監督(unsupervised)的人工神經網絡方式以協同過濾 (collaborative filtering)的形式訓練嵌入(embedding),從而從The Cancer Genome Atlas(TCGA)基因表達組數據集中提取生物相關信息。輸入樣本的 癌症類型和基因的語義含義等已知資訊均被證明保留在嵌入後的矩陣中。另 外,我們證實嵌入後的矩陣擁有生物學相關性,更以嵌入辨認了免疫檢查點 封鎖治療(immune checkpoint blockade therapy)生物標誌的候選基因。最後,我們擴展應用嵌入到深度學習網絡模型(deep learning)中,期望 直接從轉錄組學數據而非單個或數個生物標記來預測化療成效的效果。該深 度學習模型分別由藥物擾動預測器(DPP)和藥物反應預測器(DRP)兩個模 塊組組成,並分別擷取了The Library of Integrated Network-Based Cellular Signatures(LINCS)和Cancer Cell Line Encyclopedia(CCLE)中的數據訓練 DPP 和DRP 模組。該模型預測傳統具細胞毒性的化療藥物不適宜用作化療 增效劑或聯合其他藥物一併使用,反而組蛋白去乙酰酶抑製劑(HDAC inhibitor)普遍可協同其他藥物使癌細胞對治療更為敏感。癌細胞系藥物試驗 中亦確認了模型的預測。因此,我們相信深度學習可以有望成為精準癌症治 療中化療提案的重要工具。總言而之,本文探討了隨機仿真數模,嵌入和深度學習在預測化療耐藥 性、識別生物標誌和提出有效的化療提案的效用,相信這項研究能促使AI 在 腫瘤學上的應用,從而加快精準癌症治療的臨床應用,使癌症患者受益。.
590
$a
School code: 1307.
650
4
$a
Oncology.
$3
751006
690
$a
0992
710
2
$a
The Chinese University of Hong Kong (Hong Kong).
$3
1017547
773
0
$t
Dissertations Abstracts International
$g
81-09B.
790
$a
1307
791
$a
Ph.D.
792
$a
2019
793
$a
English
856
4 0
$u
http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=27783952
筆 0 讀者評論
館藏地:
全部
電子資源
出版年:
卷號:
館藏
1 筆 • 頁數 1 •
1
條碼號
典藏地名稱
館藏流通類別
資料類型
索書號
使用類型
借閱狀態
預約狀態
備註欄
附件
W9416669
電子資源
11.線上閱覽_V
電子書
EB
一般使用(Normal)
在架
0
1 筆 • 頁數 1 •
1
多媒體
評論
新增評論
分享你的心得
Export
取書館
處理中
...
變更密碼
登入