賴氨酸丙二酰化作用(Malonylation)是一種新發現的蛋白質轉譯後修飾作用(Post-translational modification, PTM)，發生該作用之蛋白質在許多生物功能中扮演著重要的角色，例如：動物蛋白質中葡萄糖及脂肪酸的代謝途徑、2型糖尿病的發病機制，和植物的碳代謝作用等。即便如此，目前對於丙二酰化作用之相關機制的研究成果仍相當有限。而其中，找出丙二酰化之作用位點為分析其相關機制的一個重要過程。傳統的驗證方式主要是透過在實驗室中的一系列生物實驗與分析，然而此方法之人力、時間與金錢成本都相當高。利用計算生物學的技術來辨識蛋白質轉譯後修飾作用位點已成為重要的研究議題。目前以計算生物學來判斷是否發生賴氨酸丙二酰化的研究大多專注於哺乳類蛋白質中，對於植物蛋白質之賴氨酸丙二酰化作用位點之預測系統卻還沒有一個專門的工具。因此，本研究提出以深度學習(Deep learning)方法來識別哺乳類與植物蛋白質之丙二酰化作用位點。我們從蛋白質胺基酸的物理化學性質、演化訊息，以及序列資訊等提取特徵，藉由混和式深度學習模型來識別發生丙二酰化作用之位置。在獨立集的測試中，分別在預測哺乳類動物蛋白質與植物蛋白質的丙二酰化作用位點模型中得到了AUC (Area under the receiver operating characteristic curve) 0.943與0.772。最後，建立了網站(Kmalo, http://fdblab.csie.ncu.edu.tw/Kmalo/)來提供這兩個預測模型。;Lysine malonylation is one of the newly recognized post-translational modification (PTMs), it is involved in many biological functions, such as cellular regulation, disease processes and carbon fixation. For better understanding the mechanisms of malonylation, identifying malonylation sites is an essential process. Traditionally, their identifications mainly rely on the mass spectrometry and biological experiments, which is time-consuming, labor-intensive and expensive. Recently, some studies have proposed computational approaches to predict malonylation sites in mammalian proteins. However, there has no predictor for malonylation sites in plant proteins. In this study, we developed two deep learning-based frameworks for identifying malonylation sites in mammalian and plant proteins separately. Physicochemical properties, evolutionary information and sequenced-based features were extracted for training the perdition models. We utilized hybrid deep learning models to predict the malonylation sites. The independent testing results for mammalian and plant proteins achieved an area under the receiver operating characteristic curve (AUC) value of 0.943 and 0.772 respectively. Furthermore, the prediction models are freely available as an online server —named Kmalo at http://fdblab.csie.ncu.edu.tw/Kmalo/.
