Research

Academic research and teaching results in recent years

In the past five years of academic results, our main research direction is to explore the application of various machine learning algorithms to the identification of protein properties in organisms, such as quaternary structure prediction [1,2], PTM binding site prediction [3], and subcellular location prediction [4]. In addition, genetic algorithm-based protein-ligand docking software has also been used to identify potential anticancer compounds for melanoma treatment [5]. In addition, an industry-academia collaboration is developing a application of convolutional neural network for image identification of the capacity of ants in their nests. In the future, the existing results will be continued and the research will be carried out in the industrial application of bioinformatics, medical informatics and AI deep learning.

We also instruct undergraduate students to conduct artificial intelligence/deep learning projects. After one year of cultivation and training, students can eventually develop the deep learning models with good predictive performance and infer it to the task of AI applications. Its application topics include medical fields, such as respiratory sounds to distinguish lung diseases, skin melanoma image recognition; And biological ecology, such as the identification of fall armyworm and larvae of lepidoptera, and the identification of common ant species in Taiwan.


  1. Chi-Hua Tung, Ching-Hsuan Chien, Chi-Wei Chen, Lan-Ying Huang, Yu-Nan Liu, Yen-Wei Chu, "QUATgo: Protein quaternary structural attributes predicted by two-stage machine learning approaches with heterogeneous feature encoding," 2020, PLoS ONE, 15(4): e0232087. (SCI, MULTIDISCIPLINARY SCIENCES, Rank: 24/69, IF: 2.776)
  2. Chi-Hua Tung, Chi-Wei Chen, Ren-Chao Guo, Hui-Fuang Ng, and Yen-Wei Chu, "QuaBingo: A Prediction System for Protein Quaternary Structure Attributes Using Block Composition," 2016, BioMed Research International, Volume 2016 (2016). (SCI, BIOTECHNOLOGY & APPLIED MICROBIOLOGY, Rank: 67/160, IF: 2.476)
  3. Chi-Chang Chang, Chi-Hua Tung, Chi-Wei Chen, Chin-Hau Tu, Yen-Wei Chu, "SUMOgo: Prediction of sumoylation sites on lysines by motif screening models and the effects of various post-translational modifications," 2018, SCIENTIFIC REPORTS, 8(2018), 15512. (SCI, MULTIDISCIPLINARY SCIENCES, Rank: 15/69, IF: 4.011)
  4. Chi-Hua Tung, Chi-Wei Chen, Han-Hao Sun, and Yen-Wei Chu, "Predicting human protein subcellular localization by heterogeneous and comprehensive approaches," 2017, PLoS ONE, 12(6): e0178832. (SCI, MULTIDISCIPLINARY SCIENCES, Rank: 15/64, IF: 2.766)
  5. Chi-Hua Tung, Yen‐Ta Lu, Wei‐Ting Kao, Jen‐Wei Liu, Yi‐Hsuan Lai, Shinn‐Jong Jiang, Hao‐Ping Chen, Tzenge‐Lien Shih, "Discovery of a more potent anticancer agent than C4‐benzazole 1,8‐naphthalimide derivatives against murine melanoma," 2020, JOURNAL OF THE CHINESE CHEMICAL SOCIETY, 67(7): 1254–1262. (SCI, CHEMISTRY, MULTIDISCIPLINARY, Rank: 128/172, IF: 1.554)

The Ministry of Science and Technology (MOST) Projects

  1. Developing disease-USA network with weighted RWR score for protein pathogenic prediction

    • 2016/08/01 - 2017/07/31

    • Funding: NTD 423,000

  2. Structure-Oriented Distance Motif for Protein Functions Prediction and Annotation

    • 2012/08/01 - 2013/07/31

    • Funding: NTD 528,000

  3. Networks of Protein Structural Units in Pharmaceutical Applications

    • 2012/01/01 - 2012/07/31

    • Funding: NTD 500,000

Industry-academia Cooperation Project

  1. 整合影像辨識分析螞蟻成長過程-螞蟻生態日記服務平台 (Integrated Image Recognition and Analysis of Ant Growth: An Ant Ecological Diary Service Platform)

Protein Structural Units Networks

Using SA, we define new local structural fragments called units of structural alphabet (USAs) that represent unique features of protein structures. Each USA is composed of two secondary protein structures and one loop located between these two secondary structures. USAs can maintain not only the flexibility of variable loops but also the stability of secondary structures. We conduct a similarity search and investigate the network formed by all-against-all USA sequence comparisons, where USAs represent nodes and links represent homology relationships.

Our findings show a highly uneven degree distribution characterized by a few and highly connected USAs (hubs) coexisting with many nodes having only a few links. Networks with such a power-law degree distribution are scale free. These findings not only suggest the existence of organizing principles for local protein structures but also allow us to identify key fragments that are potentially useful for new drug development and design. Of particular interest is the identification of USAs in the set of known drug protein targets.

3D-BLAST Protein Structure Search

We developed the kappa-alpha (κ, α) plot derived structural alphabet and a novel BLOSUM-like substitution matrix, called structural alphabet substitution matrix, which searches through the structural alphabet database. This structural alphabet (SA) was used in developing the fast structure database search method called 3D-BLAST, which is as fast as BLAST and provides the statistical significance (E-value) of an alignment, indicating the reliability of a hit protein structure. Moreover, we developed an automated server called fastSCOP for integrating a fast structure database search tool (3D-BLAST) and a detailed structural comparison tool, as well as for recognizing the SCOP domains and SCOP superfamilies of query structures.