治療性肽(therapeutic peptide)指的是具有醫療功能的短序列蛋白質，而使用到治療性肽的療程就被稱之為胜肽療法(peptide therapy)。胜肽療法因為比起傳統療法來說具有對人體低毒性、容易被人體分解、高選擇性等等多種好處，已是現代生醫領域中的不可忽視的研究重點之一。
治療性肽的功能非常多樣，現今已被證實的功能就包括了抗微生物、抗寄生蟲、抗腫瘤、抗真菌、抗病毒、抗細菌等多種反應目標，足以顯示其種類之廣泛，且一條肽鍊可同時具備多種反應對象，如抗葛蘭氏陽性菌肽常常同時具有抗葛蘭氏陰性菌的能力，精準辨認治療性肽的功能是胜肽療法中一個重要的課題。
傳統上一般透過生物實驗的方式來驗證肽鏈的功能，但生物實驗較為昂貴且耗時，在機器學習技術蓬勃發展的今日，許多研究都嘗試了比較快速的傳統機器學習或深度學習來建構肽鏈的分類模型，也得到了不錯的表現，證實了透過機器學習方法來分類治療性肽是可行的。
在蒐集並整理相關研究後，發現多數研究方法是獨立對每種治療性肽進行分類，較少考慮胜肽種類之間是否具備關聯性。即使有，多數也僅視為同一層級的胜肽群使用。但實際上，例如抗菌肽我們可以依照其對象進一步細分為抗葛蘭氏陽性菌和抗葛蘭氏陰性菌兩種類。若以階層式方式設計，不僅可以讓模型學習不同層級問題的知識，還可以簡化問題並幫助模型學習。
本研究建立了一個階層式的卷積神經網路分類器。首先模型將學習如何判斷肽鏈是否具有治療功能的高層級知識，此時分類標準為假定的治療性肽，故稱之為偽標籤，後續再進一步去學習分辨特定的治療功能，例如抗腫瘤或抗病毒，最後再混合前面兩層級所學出來的特徵以強化分類表現。在與相關研究比較中，本研究所提出的模型在多種治療性肽的馬修斯相關係數得到了更好的表現。

Therapeutic peptides are short protein sequences that have the ability to fight off illnesses. The term "therapeutic peptides" is derived from the concept of "peptide therapies," which refers to treatments utilizing these peptides. Compared to traditional treatments, peptide therapies offer many benefits, such as low toxicity, easier metabolism by the human body, and high selectivity. These advantages make peptide therapies indispensable in medical developments.
Before any peptide is deployed for practical usage, we must first verify its targets and mechanisms. Many kinds of therapeutic peptides have been discovered, including anti-microbial peptides, anti-parasite peptides, anti-cancer peptides, anti-fungal peptides, anti-virus peptides, and anti-bacterial peptides. Additionally, a peptide sequence can have multiple reactions toward various target cells. For example, a peptide that is effective against Gram-positive bacteria is likely to also have an effect on Gram-negative bacteria. Therefore, efficiently classifying these peptides' functions is critical.
Traditionally, we use biological experiments to test peptide sequences, but these experiments are costly and inefficient. Nowadays, many researchers have introduced computational methods into peptide classification, and the results indicate that these computational approaches can yield incredible results.
After collecting and organizing relative works, it was found that most research methods classify each type of therapeutic peptide independently, with little consideration of potential relationships between different peptide types. Even in cases where such relationships are considered, such as in methods like MLBP, they are often treated as belonging to the same hierarchical level. In practice, however, anti-bacterial peptides can be further subdivided based on their targets into categories like anti-Gram-positive and anti-Gram-negative peptides. Designing a hierarchical approach allows the model to learn knowledge at different levels of granularity, simplifying the problem and aiding in model learning.
This study establishes a hierarchical convolutional neural network classifier. Initially, the model learns high-level knowledge to determine whether a peptide sequence has therapeutic functions, using a pseudo label since the classification criterion does not encompass all real therapeutic peptides. Subsequently, it delves deeper to learn specific therapeutic functions, such as anti-cancer or anti-viral properties. Finally, the features learned at these two levels are combined to enhance classification performance. Compared with related studies, the proposed model achieves better Matthews Correlation Coefficient (MCC) performance across various therapeutic peptides.

[1] P. Bhadra, J. L. Yan, J. Y. Li, S. Fong, and S. W. I. Siu, "AmPEP: Sequence-based prediction of antimicrobial peptides using distribution patterns of amino acid properties and random forest," (in English), Sci Rep-Uk, vol. 8, Jan 26 2018, doi: ARTN 169710.1038/s41598-018-19752-w.
[2] B. Manavalan, S. Basith, T. H. Shin, S. Choi, M. O. Kim, and G. Lee, "MLACP: machine-learning-based prediction of anticancer peptides," (in English), Oncotarget, vol. 8, no. 44, pp. 77121-77136, Sep 29 2017, doi: 10.18632/oncotarget.20365.
[3] W. Chen, H. Ding, P. M. Feng, H. Lin, and K. C. Chou, "IACP: a sequence-based tool for identifying anticancer peptides," (in English), Oncotarget, vol. 7, no. 13, pp. 16895-16909, Mar 29 2016, doi: DOI 10.18632/oncotarget.7815.
[4] X. Xiao, P. Wang, W. Z. Lin, J. H. Jia, and K. C. Chou, "iAMP-2L: A two-level multi-label classifier for identifying antimicrobial peptides and their functional types," (in English), Anal Biochem, vol. 436, no. 2, pp. 168-177, May 15 2013, doi: 10.1016/j.ab.2013.01.019.
[5] S. Lata, N. K. Mishra, and G. P. S. Raghava, "AntiBP2: improved version of antibacterial peptide prediction," (in English), Bmc Bioinformatics, vol. 11, 2010, doi: Artn S1910.1186/1471-2105-11-S1-S19.
[6] C. R. Chung, T. R. Kuo, L. C. Wu, T. Y. Lee, and J. T. Horng, "Characterization and identification of antimicrobial peptides with different functional activities," (in English), Brief Bioinform, vol. 21, no. 3, pp. 1098-1114, May 2020, doi: 10.1093/bib/bbz043.
[7] S. L. Zhang and X. J. Li, "Pep-CNN: An improved convolutional neural network for predicting therapeutic peptides," (in English), Chemometr Intell Lab, vol. 221, Feb 15 2022, doi: ARTN 10449010.1016/j.chemolab.2022.104490.
[8] W. D. Tang et al., "Identifying multi-functional bioactive peptide functions using multi-label deep learning," (in English), Brief Bioinform, vol. 23, no. 1, Jan 17 2022, doi: ARTN bbab41410.1093/bib/bbab414.
[9] Y. Li, X. Y. Li, Y. W. Liu, Y. H. Yao, and G. H. Huang, "MPMABP: A CNN and Bi-LSTM-Based Method for Predicting Multi-Activities of Bioactive Peptides," (in English), Pharmaceuticals-Base, vol. 15, no. 6, Jun 2022, doi: ARTN 70710.3390/ph15060707.
[10] 沈柏妤, "使用異質卷積神經網路預測抗微生物肽," 碩士, 電機工程學系, 國立成功大學, 台南市, 2020. [Online]. Available: https://hdl.handle.net/11296/c54mqq
[11] G. F. Dong, L. Zheng, S. H. Huang, J. Gao, and Y. C. Zuo, "Amino Acid Reduction Can Help to Improve the Identification of Antimicrobial Peptides and Their Functional Activities," (in English), Front Genet, vol. 12, Apr 20 2021, doi: ARTN 66932810.3389/fgene.2021.669328.
[12] Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-based learning applied to document recognition," (in English), P Ieee, vol. 86, no. 11, pp. 2278-2324, Nov 1998, doi: Doi 10.1109/5.726791.
[13] H. Kaiming, Z. Xiangyu, R. Shaoqing, and S. Jian, "Deep Residual Learning for Image Recognition," 2016: IEEE, pp. 770-778, doi: 10.1109/CVPR.2016.90.
[14] G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, "Densely Connected Convolutional Networks," 2017: IEEE, pp. 2261-2269, doi:10.1109/CVPR.2017.243.
[15] J. Hu, L. Shen, S. Albanie, G. Sun, and E. H. Wu, "Squeeze-and-Excitation Networks," IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, vol. 42, no. 8, pp. 2011-2023, AUG. 1 2020, doi:10.1109/TPAMI.2019.2913372.