整合急性傳染性疾病，如當前新型冠狀病毒疾病（COVID-19）的相關資訊以尋求預防與治療的方法是相當急迫且重要的議題。一種可能的作法是透過分析最新的學術文獻相關資料，取得有用的資訊。這些文獻是來自世界各地專家的最新研究，透過挖掘相關文章找出可能與新型冠狀病毒疾病有關的資訊。然而目前單是PubMed上就有超過3000萬篇生物醫學文獻，預計未來甚至會更快速增加，如何快速精準的處理此巨量資料是另一個問題。隨著人工智慧的發展，應用文字探勘與機器閱讀工具可能得以解決此問題。
本研究的目的是整合醫學文獻，運用人工智慧技術建立一套新型冠狀病毒疾病(COVID-19)特定的文獻探勘系統。我們首先建立網路爬蟲，使用由 (1) 病毒造成的人類疾病和中草藥化合物關鍵字；(2) 病毒造成的人類疾病和印度藥用植物關鍵字等兩組查詢語法，定期自動的從PubMed生物醫學文獻庫中檢索與新型冠狀病毒疾病相關的文獻；透過文字探勘的類別辨識(Named entity recognition, NER)子系統辨識收集的文獻，標記出文獻中的七種(基因/蛋白質、疾病、藥物/化合物、物種、突變、小分子核糖核酸、作用路徑)類別名稱，構建COVID-19的特定資料庫；再透過關係擷取(Relation extraction, RE)子系統取得文獻內基因與疾病之間的關聯，進而建構基因與疾病之網路關聯圖。再者，我們擷取BioGRID資料庫中COVID-19的30種病毒蛋白與人類基因之間交互作用的資料，建立人類基因及病毒基因的關聯網路，並且透過圖形理論的網路分支度性質找尋受到最多病毒基因影響的人類基因。在此關聯網路中網路分支度較高的人類蛋白如ATP1A1(網路分支度為20)及AP2M1(網路分支度為16)等，均有文獻證明它們的確在COVID-19疾病中扮演重要的角色。此外，S蛋白(Spike protein)是COVID-19主要進入宿主的病毒蛋白，本研究根據所建立的人類基因及病毒基因關聯網路，篩選與S蛋白直接作用的人類基因，並透過功能註釋工具DAVID找出這些人類基因群集潛在的生物功能。
本研究所建立之COVID-19文獻探勘系統包含類別識別子系統與關係擷取子系統，未來可進一步運用於其他生物醫學文本建立有效的資訊檢索系統。

Integrating the related information in order to prevent and cure the infectious diseases, especially the current COVID-19, is an very urgent and important issue. One possible solution is to find useful information to construct prediction models by analyzing the state-of-the-art medical publications. As these literatures are studies from experts around the world, there are huge potential to find useful drugs and compounds for diseases treatments. However, there are vast volume of text materials, literature, and publications that are related to this field. For instance, there are over than 30 millions of biomedical literature deposited in PubMed alone and they are expected to grow even rapidly in the future. How to fast and accurate identification for the related information is another problem. With the development of artificial intelligence, applying text mining tools would reduce time and effort to find and extract useful information from these and machine reading tools could possibly solve this problem.
This study is to construct the text mining system for COVID-19 by artificial intelligence technology. We started by developing web crawler program, which regularly and automatically fetch the related medical literature for COVID-19 from PubMed. Two kinds of query strings: (1) keywords based on diseases caused by virus (SARS, Coronavirus、HIV、MERS and Ebola) along with Indian pharmaceutical plants (2) keywords based on diseases caused by virus (SARS, Coronavirus, HIV, MERS and Ebola) along with Chinese medicinal herbs were used for searching in the PubMed. And then, we constructed the name entity recognition (NER) subsystem for locating and classifying of entities in literatures into seven categories, including gene/protein, disease, drug/chemicals, species, mutations, miRNAs and pathways. With the NER subsystem, we developed a specified database for COVID-19. We also constructed the relation extraction (RE) subsystem to derive the relationships among gene and disease in literatures, by which, the corresponding associated networks were constructed for further analysis by graph theory. Furthermore, we derived the interactions among 30 viral proteins of COVID-19 and human genes from the BioGRID database, established networks of associations between human genes and viral genes. And then, we searched for human genes that were most affected by the virus genes through the network degree centrality property. We found that human proteins with high network degree in this associated network, such as ATP1A1 (network degree is 20) and AP2M1 (network degree is 16), both have been well documented to be significantly involved in COVID-19 disease. In addition, spike protein is a crucial protein that mainly enters the host of COVID-19. Based on the established network of human genes and viral gene associations, this study screened human genes that directly interacted with S proteins and identified the potential biological functions of these human gene clusters through the functional annotation tool DAVID.
The COVID-19 literature mining system established by this study contains name recognition subsystems and relationship extraction subsystems, and the system could be used to establish effective information retrieval systems for other biomedical texts.

中文摘要 …………………………………………………………i
英文摘要 …………………………………………………………iii
誌謝 …………………………………………………………v
目錄 …………………………………………………………vi
表目錄 …………………………………………………………viii
圖目錄 …………………………………………………………ix
第一章 背景、動機與目的………………………………………1
1.1 動機與目的…………………………………1
1.2 人工智慧………………………………………………1
1.3 文字探勘與自然語言處理……………………………2
1.4 COVID研究現況………………………………………2
第二章 研究方法………………………………………………3
2.1 研究流程………………………………………………3
2.2 原始資料………………………………………………3
2.3 前處理…………………………………………………6
2.4 語言代表模型BERT……………………………………7
2.5 生物醫學自然語言處理模型BioBERT………………7
2.6 類別辨識子系統 NER…………………………………7
2.7 關係擷取子系統 RE……………………………………9
2.8 透過BioGRID資料庫建立人類基因及病毒基因的關聯網路…………………………………………………11
2.9 網路中心性指標Degree centrality……………………14
2.10 以DAVID工具分析關鍵基因Spike protein關聯的人類基因………………………………………14
2.11 使用圖形理論分析病毒基因與人類基因關聯網路……15
第三章 研究結果………………………………………17
3.1 使用NER子系統建立七種類別的資料庫……………17
3.2 使用BioGRID資料庫建立人類基因及病毒基因關聯網路……………………………………………………18
3.3 使用BioGRID建立人類基因及人類基因交互作用網路………………………………………………………23
3.4 計算人類個別基因受病毒基因影響關係的強弱……23
3.5 處理關鍵基因集……………………………………23
3.6 使用DAVID工具分析與S蛋白關聯的人類基因……24
3.7 RE子系統建立基因與疾病關聯網路…………………27
第四章 結論……………………………………………………31
參考文獻 …………………………………………………………32
英文論文大綱…………………………………………………………35

[1] Shaw, D., Chen, H. and Jiang, T., 2018, “A deep domain adaptation approach to predict isoform functions”, Bioinformatics, 35(15), 2535-2544, August.
[2] Stahl, K., Schneider, M. and Brock, O., 2017, “EPSILON-CP: using deep learning to combine information from multiple sources for protein contact prediction”, BMC Bioinformatics, 18(303), June.
[3] Skarga-Bandurova, I., Biloborodova, T. and Nesterov, M., 2018, “Extracting interesting rules from gestation course data for early diagnosis of neonatal hypoxia”, Journal of medical systems, 43(8), December.
[4] Moradi, M., 2018, “CIBS: A biomedical text summarizer using topic-based sentence clustering”, Journal of biomedical informatics, 88, 53-61, December.
[5] Zhou, J. and Fu, B.Q., 2018, “The research on gene-disease association based on text-mining of PubMed”, BMC Bioinformatics, 19(1), 37, February.
[6] Poofery, J., et al., 2020, “Potential of Thai herbal extracts on lung cancer treatment by inducing apoptosis and synergizing chemotherapy”, Molecules, 25(1), 231, January.
[7] Lin C.J., et al., 2019, “High amplification of the antiviral activity of curcumin through transformation into carbon quantum dots”, Small, 15(41), e1902641, October.
[8] Liu Y.M., et al., 2020, “A carbohydrate-binding protein from the edible Lablab beans effectively blocks the infections of influenza viruses and SARS-CoV-2”, Cell Reports, 32(6), 108016, August.
[9] Jumper, J., Tunyasuvunakool, K., Kohli, P., Hassabis, D., and the AlphaFold team., 2020, “Computational predictions of protein structures associated with COVID-19”, DeepMind website, March, https://deepmind.com/research/open-source/computational-predictions-of-protein-structures-associated-with-COVID-19.
[10] Ko, H., et al., 2020, “COVID-19 Pneumonia Diagnosis Using a Simple 2D Deep Learning Framework With a Single Chest CT Image: Model Development and Validation”, Journal of medical Internet research, 22(6), e19569, June.
[11] Osakwe, Z.T., et al., 2021, “Identifying public concerns and reactions during the COVID-19 pandemic on Twitter: A text-mining analysis”, Public health nursing (Boston, Mass.), 38(2), 145-151, November.
[12] Singh, A.K. and Khunti, K., 2020, “Assessment of risk, severity, mortality, glycemic control and antidiabetic agents in patients with diabetes and COVID-19: A narrative review”. Diabetes research and clinical practice, 165, 108266, July.
[13] Singh, R., Kaur, N., Kishore, L. and Gupta, G. K., 2013, “Management of diabetic complications: A chemical constituents based approach”. Journal of Ethnopharmacology, 150, 51-70, October.
[14] Noolu, B., et al., 2013, “Murraya koenigii leaf extract inhibits proteasome activity and induces cell death in breast cancer cells”, BMC complementary and alternative medicine, 13, 7, January.
[15] Gupta, P., Nahata, A. and Dixit, V. K., 2011, “An update on Murraya koenigii spreng: a multifunctional Ayurvedic herb”, Zhong Xi Yi Jie He Xue Bao=Journal of Chinese integrative medicine, 9(8), 824-33, August.
[16] Dragos, D. and Gilca, M., 2018, “PhytoMolecularTasteDB: An integrative database on the "molecular taste" of Indian medicinal plants”, Data in brief, 19, 1237-1241, August.
[17] Yang Y., et al., 2020, “Traditional Chinese Medicine in the Treatment of Patients Infected with 2019-New Coronavirus (SARS-CoV-2): A Review and Perspective”, International Journal of Biological Sciences, 16(10), 1708-1717, March.
[18] Huang, Y. F., et al., 2020, “Review on the potential action mechanisms of Chinese medicines in treating Coronavirus Disease 2019 (COVID-19)”, Pharmacol Research, 158, 104939. May.
[19] Devlin, J., Chang, M., Lee, K. and Toutanova, K., 2019, “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding”, NAACL-HLT, 1, 4171-4186, June.
[20] Vaswani, A., et al., 2017, “Attention is All you Need”, Advances in Neural Information Processing Systems, 5998-6008, December.
[21] Lee, J., et al., 2020, “BioBERT: a pre-trained biomedical language representation model for biomedical text mining”, Bioinformatics, 36(4), 1234-1240, February.
[22] Kim D., et al., 2019, “A Neural Named Entity Recognition and Multi-Type Normalization Tool for Biomedical Text Mining,”, IEEE Access, 7, 73729-73740, June.
[23] Chatr-Aryamontri, A., et al., 2013, “The BioGRID interaction database: 2013 update”, Nucleic Acids Research, 41(Database issue), D816-D823, January.
[24] 陳登鴻，2015，“蛋白質複合體交互作用網路之拓樸分析”。國立虎尾科技大學資訊工程研究所碩士論文。
[25] Rawat, K., Kumari, P. and Saha, L., 2021, “COVID-19 vaccine: A recent update in pipeline vaccines, their design and development strategies”. European journal of pharmacology, 892, 173751, February.
[26] Huang, W., Sherman, B.T. and Lempicki, R.A., 2009, “Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources”, Nature Protocols, 4, 44-57, December.
[27] Floyd, R.W., 1962, “Algorithm 97: Shortest path”. Communications of the ACM, 5, 345, June.
[28] Yuan, S., et al., 2020, “Viruses harness YxxØ motif to interact with host AP2M1 for replication: A vulnerable broad-spectrum antiviral target”, Science advances, 6(35), eaba7910, August.
[29] Wang, P.G., et al., 2020, “Sunitinib reduces the infection of SARS-CoV, MERS-CoV and SARS-CoV-2 partially by inhibiting AP2M1 phosphorylation”, Cell discovery, 6(71), October.
[30] Schmidt, N., et al., 2021, “The SARS-CoV-2 RNA–protein interactome in infected human cells”, Nat Microbiol, 6, 339-353, March.
[31] Mohammed, M., 2021, “SARS-CoV-2 proteins: Are they useful as targets for COVID-19 drugs and vaccines?”, Current molecular medicine, 21, https://doi.org/10.2174/1566524021666210223143243 , February.
[32] Shannon, P., et al., 2003, “Cytoscape: a software environment for integrated models of biomolecular interaction networks”. Genome research, 13(11), 2498-2504, November.