生物標記 (biomarker) 泛指可以利用任何生物分析方式來測量正常人體生理代謝、疾病發展過程及藥物治療過程等，同時可以提供臨床上決策依據的生物指標，協助預防及治療。尋找生物標記的過程是十分困難的，微陣列 (microarray) 晶片的推出，幫助我們尋找生物標記物更為方便及快速。
本研究便是使用微陣列晶片之一的大腸桿桿菌蛋白質微陣列 (proteome microarray) 晶片做為分析資料集，且利用它來找到生物標記。透過掃描器掃描完成後的晶片圖檔，一般會使用微陣列影像分析軟體來讀取。在本研究中使用的是 GenePix，經過人工對點後，該軟體會自動偵測出晶片上各點的表現量 (expression)，而給出各個蛋白質前景 (foreground) 及背景 (background) 的數值，之後利用這些數據篩選出有差異表現的蛋白質。但我們發現 GenePix 軟體輸出的表現量，常常發生數值與肉眼實際亮度不一致的情形，導致找到的潛在生物標記失真。
為此，本研究提出了一個基於深度學習的演算法，可以給出更準確的蛋白質表現量，該數值有助於在後續辨識差異蛋白質分析上，且可以找到與利用微陣列影像分析軟體給出的表現量不一樣的蛋白質。深度學習模型有效的解決了亮度不一致的情況，也基於此演算法找到許多不只與肉眼吻合的候選蛋白質。

Biomarker, an indicator that is able to measure metabolism, the process of development of disease, treatment, and at the same time be able to help making clinical decision. As it is difficult to finding a biomarker, microarray offers a more effective and more convenient way to find biomarkers.
The E. coli proteome microarray, one of the microarrays, is used as the analysis dataset in this study, and it is used to find biomarkers. The chip image file produced by the scanner is generally read by microarray image analysis software, in this work is GenePix. The software automatically detects the signal intensity of each point on the chip and gives the value of the foreground and background of each protein after manual alignment. The following analysis algorithms then identify the difference in performance to select proteins with differential expression based on these data. However, inconsistency between the intensity of the signal given by GenePix and the actual brightness perceived by naked eye is found in this work, which results in distortion of the result of selected potential biomarkers.
In this study, we aim to propose a deep-learning-based algorithm to give more accurate gene expressions. This deep learning model effectively solves the inconsistency of brightness, and the results are verified on the Kawasaki disease dataset. we found many candidate proteins that are not only compatible with the naked eye. In addition, combining candidate proteins can achieve higher efficacy (AUC) base on such algorithm.

1. Instruments, A., GenePix Pro 6.0 Microarray acquisition and analysis software for GenePix microarray scanners User Guid and Tutorial. 2005.
2. Tusher, V.G., R. Tibshirani, and G. Chu, Significance analysis of microarrays applied to the ionizing radiation response. Proceedings of the National Academy of Sciences, 2001. 98(9): p. 5116-5121.
3. MikeJSeo, SAM (Significance Analysis of Microarrays) shiny app https://github.com/MikeJSeo/SAM.
4. He, K., et al. Deep residual learning for image recognition. in Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.
5. Wright, R.E., Logistic regression. 1995.
6. Kuo, H.-C., et al., Antibody profiling of Kawasaki Disease using Escherichia coli proteome microarrays. Molecular & Cellular Proteomics, 2018. 17(3): p. 472-481.
7. Pizer, S.M., et al., Adaptive histogram equalization and its variations. Computer vision, graphics, and image processing, 1987. 39(3): p. 355-368.