在世代研究中發現到有下泌尿道發炎或感染的男性患者，罹患前列腺癌的風險較常人高出許多。並且，已有報導指出前列腺感染會促進前列腺癌發生的機率，而大部分的前列腺癌感染又與大腸桿菌有關。因此我們使用大腸桿菌蛋白質體晶片來分析前列腺癌患者血液中抗體的資訊。找出臨床上癌症顯著患者與臨床上癌症顯著患者血液中抗體之大腸桿菌抗原的差異，以期建立前列腺癌分期的診斷工具。首先我們將大腸桿菌蛋白質體固定在玻璃晶片上,進行實驗時先使用 3% 胎牛血清白蛋白填滿晶片表面未附著蛋白質處,以防非特異性結合發生。接下來加入 2000倍稀釋的受試者血清,再將其與晶片上大腸桿菌蛋白反應。再利用 PBST洗去所有未結合的抗體及其他雜質。最後使用標有螢光Dylight 550的抗人體抗體IgG Dylight 650的抗人體抗體IgM來標示受試者抗體。同樣使用 PBST 洗去未結合的 IgG 以及 IgM, 利用掃描器偵測螢光以及查出對應位置的蛋白質名稱, 再比對不同組別之間抗體的種類差異並進行統計。統計出的結果將會用另一批患者的血清來進行驗證。在結果分析發現到，良性組、中危組和高危組三組數據顯示無顯著性差異。我們只找到4個具有高於70%的準確度的蛋白質顯示訊號在中危組高於良性組。在高危組與中危組之間以及在高危組與良性組之間並未發現任何蛋白具有高60%的準確度。所以我們將中危組以及高危組合併為中高危組，與良性組樣本進行比較。並且將總樣本分為訓練資料集和驗證資料集。首先，我們處理訓練資料集樣本後，我們使用AI模型讀取信號強度。篩選出組間信號差異大（p值<0.05）的蛋白質，此步驟中獲得約600個蛋白質。從中使用人眼再次篩選並選擇超過 70% 準確度的蛋白質17個和超過75% 靈敏度的蛋白質19個和超過75% 特異性的蛋白質20個。這些蛋白質再用驗證資料集進一步驗證。用人眼再次篩選凡準確度、靈敏度或特異性大於七成者。此步驟中篩選出了1個蛋白質(ynfM)具有高於70 %的準確率 (70.37 %)，6 個蛋白質具有高於七成的靈敏度和14個蛋白質具有高於七成的特異性。接下來我們針對這些21個在驗證資料集中通過準確度、靈敏度或特異性大於七成驗證的蛋白質進行邏輯回歸分析。結果從AI 模型數據計算分析中，只得到 ynfM (IgG) 和 ybaY(IgG) 的組合在訓練資料集和驗證資料集的準確率皆高於七成，此組合的準確率 (訓練資料集: 81.1% 驗證資料集: 72.2%) 略高於只有ynfM 的準確率(訓練資料集: 71.7% 驗證資料集: 70.37%)。在傑出規則讀取數據分析中，我們發現只有 yrbC(IgM)、dusB(IgG)、mngA(IgG)組合在訓練資料集和驗證資料集的準確率皆高於七成(訓練資料集: 84.9%, 驗證資料集: 70.4%)。這些結果有助於預測前列腺腫瘤患者的預後以及是否需要進行切除手術，減少對患者不必要的傷害。

In some cohort studies, it has been found that male patients with lower urinary tract inflammation or infections have a higher risk of developing prostate cancer. Besides, it has been reported that prostate infections promote the occurrence of prostate cancer, and most prostate infections are related to Escherichia coli. Thus, we used E.coli proteome chips to analyze the antibody profile of prostate cancer patient serum. By identifying the antibody difference between clinically insignificant prostate cancer patients and clinically significant prostate cancer patients, we may use the antibody as a prognosis marker for prostate cancer operations. First, we printed ~4800 E. coli proteins on Aldehyde coated slides. In the assay experiment, we used 3% bovine serum albumin for blocking to prevent non-specific binding. Next, we added 2000-fold dilution of the patient's serum to make the antibody bind to the E. coli protein. Then we used PBST to wash away all unbound antibodies and other compounds in the serum. Finally, fluorescent Dylight 550 labeled anti-human IgG and fluorescent Dylight 650 labeled anti-human IgG were used to indicate patient’s antibodies on the chip. Also, we used PBST to wash away unbound fluorescent labeled antibody IgG and IgM, and used a fluorescent laser scanner to detect fluorescence and identify the protein name according to the corresponding location. Then, we compared the differences between different groups in the training set. The proteins which showed differences were further validated with another batch of patient’s serum (validation set).The results showed that the antibody profiles among the benign, moderate and high risk group were not significantly different. We only found four proteins have over 70% accuracy, which have higher signal in moderate than benign group. High and moderate as well as high and benign group did not have any proteins more than 60% accuracy. Thus, we combined the moderate group and high risk group sample as moderate+high group (significant cancer) to compare with the benign group (insignificant cancer) sample. First, we worked on the training set samples. We used the AI model to read the signal strength. We selected proteins which shows significant signal difference between groups (p value < 0.05) and got about 600 proteins. From these results we screened again by eyeballing and chose accuracy or sensitivity or specificity over 70%. We got 17 proteins over 70 % of accuracy, 19 proteins over 75% of sensitivity and 20 proteins over 75% of specificity. These proteins are further verified by the validation data set to select proteins that the accuracy, sensitivity or specificity is greater than 70% by eyeballing. In this step, 1 protein (ynfM) was screened with accuracy rate higher than 70% (70.37%), 6 proteins had sensitivity higher than 70%, and 14 proteins had specificity higher than 70%. Next, we performed logistic regression analysis for these 21 proteins whose accuracy, sensitivity, or specificity is greater than 70% in the validation data set. For the analysis from the AI model data, only the combination of ynfM (IgG) and ybaY (IgG) has an accuracy rate higher than 70% both in validation data set and training data set. The accuracy of this combination (training data set: 81.1%, validation data set: 72.2%) is slightly higher than the accuracy rate of only ynfM (training data set: 71.7%, validation data set: 70.37%). In the analysis of the outstanding rule model data, we found that only the combination of yrbC (IgM), dusB (IgG), and mngA (IgG) has an accuracy rate higher than 70% both in the validation data set and the training data set (training data set: 84.9% , validation data set: 70.4%). These results may help predicting the prognosis of prostate cancers patients and the need for surgery, reducing unnecessary harm to patients.

1. Liu K, Wang A, Ran L, Zhang W, Jing S, Wang Y, Zhang X, Liu G, Wang S & Song F.2019. ARHGEF38 as a novel biomarker to predict aggressive prostate cancer. Genes & Diseases 7, 217e224 https://doi.org/10.1016/j.gendis.2019.03.004

2. Lin Y, Zhao X, Miao Z, Ling Z, Wei X, Pu J, Hou J & Shen B.2020. Data‑driven translational prostate cancer research: from biomarker discovery to clinical decision Lin et al. J Transl Med (2020) 18:119 https://doi.org/10.1186/s12967-020-02281-4

3. RICHARD G. S, M.D., JAMIE A. C, M.D.& NELSON N. S, M.D. (2005) DISEASE-SPECIFIC SURVIVAL FOLLOWING THE BRACHYTHERAPY MANAGEMENT OF PROSTATE CANCER Int. J. Radiation Oncology Biol. Phys., Vol. 64, No. 3, pp. 810–816, 2006 doi:10.1016/j.ijrobp.2005.09.005

4. Feng Y, Chen CS, Ho J, Pearce D, Hu S, Wang B, Desai P, Kwang SK, and Heng Zhu, 2018,High-Throughput Chip Assay for Investigating Escherichia coli Interaction with the Blood−Brain Barrier Using Microbial and Human Proteome Microarrays (Dual-Microarray Technology), Anal. Chem. 2018, 90, 10958−10966 ,DOI: 10.1021/acs.analchem.8b02513

5. Chen YW, Teng CH, Ho YH, Yu T, Ho J, Huang WC, Masayuki H, Chiang IY, and Chen CS. (2014) Identification of Bacterial Factors Involved in Type 1 Fimbria Expression using an Escherichia coli K12 Proteome Chip. Molecular &Cellular Proteomics 13: 10.1074/mcp.M113.035667, 1485–1494, 2014. DOI 10.1074/mcp.M113.035667

6. Tu YH, Ho YH, Chuang YC., Chen PC, Chen CS .2011.Identification of Lactoferricin B Intracellular Targets Using an Escherichia coli Proteome Chip. PLoS ONE 6(12): e28197. doi:10.1371/journal.pone.0028197

7. Chen CS, Sean S, Troy A, Aik C. Alex, Steven R. B, Carmen C, Theodore M. Bayless,Monica V. Talor, C. Lynne B, Wang H, Li R, Lisa Wu Datta, Wu Y, Raimond L. W, Zhu H, and Li X,.2009. Identification of Novel Serological Biomarkers for Inflammatory Bowel Disease Using Escherichia coli Proteome Chip. Molecular & Cellular Proteomics 8:1765–1776, 2009.DOI 10.1074/mcp.M800593-MCP200

8. Angelo M. De Marzo, Elizabeth A. P, Siobhan S, Xu J, Henrik G, Charles G. D, Yasutomo N, William B. I and William G. N, .2007. Inflammation in prostate carcinogenesis NATURE REVIEWS | CANCER VOLUME 7 | APRIL 2007

9. Karen S. S., William B. I., Angelo M De M., .2013 .Infections and inflammation in prostate cancer,Am J Clin Exp Urol 2013;1(1):3-11
10. De Marzo, A. M., et al. (2007). "Inflammation in prostate carcinogenesis." Nat Rev Cancer 7(4): 256-269.
11. Fan, C. Y., et al. (2017). "Lower Urinary Tract Infection and Subsequent Risk of Prostate Cancer: A Nationwide Population-Based Cohort Study." Plos One 12(1): 12.

12. Nordstrom, L., et al. (2013). "Foodborne urinary tract infections: a new paradigm for antimicrobial-resistant foodborne illness." Frontiers in Microbiology 4: 6.

13. Owrangi, B., et al. (2018). "Invasion and translocation of uropathogenic Escherichia coli isolated from urosepsis and patients with community-acquired urinary tract infection." Eur J Clin Microbiol Infect Dis 37(5): 833-839.

14. Udomsom, S., et al. (2021). "Novel Rapid Protein Coating Technique for Silicon Photonic Biosensor to Improve Surface Morphology and Increase Bioreceptor Density." Coatings 11(5): 19.

15. Zhuge, X. K., et al. "Chicken-sourceEscherichia coliwithin phylogroup F shares virulence genotypes and is closely related to extraintestinal pathogenic E. Coli causing human infections." Transboundary and Emerging Diseases: 16.

16. Sfanos, K. S., et al. (2015). "Bacterial Prostatitis Enhances 2-Amino-1-Methyl-6-Phenylimidazo[4,5-<em>b</em>]Pyridine (PhIP)–Induced Cancer at Multiple Sites." Cancer Prevention Research 8(8): 683-692.

17. Sfanos, K. S., et al. (2008). "A molecular analysis of prokaryotic and viral DNA sequences in prostate tissue from patients with prostate cancer indicates the presence of multiple and diverse microorganisms." The Prostate 68(3): 306-320.

18. Shaw, G. L., et al. (2014). "Identification of pathologically insignificant prostate cancer is not accurate in unscreened men." Br J Cancer 110(10): 2405-2411.

19. Krieger, J. N. and P. Thumbikat (2016). "Bacterial Prostatitis: Bacterial Virulence, Clinical Outcomes, and New Directions." Microbiol Spectr 4(1).

20. Vincent, C., et al. (2010). "Food reservoir for Escherichia coli causing urinary tract infections." Emerg Infect Dis 16(1): 88-95.

21. Goo, Y. A. and D. R. Goodlett (2010). "Advances in proteomic prostate cancer biomarker discovery." Journal of Proteomics 73(10): 1839-1850.

22. Sha, S., et al. (2020). "The human gastrointestinal microbiota and prostate cancer development and treatment." Investig Clin Urol 61(Suppl 1): S43-s50.