IMADS 2.0: an extended iMADS for MADS-box gene classification in orchid by machine learning approach

• Main Authors: Kuan-Chun Chen, 陳冠群
• Other Authors: Yen-Wei Chu
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/91440902228590407851

碩士 === 國立中興大學 === 基因體暨生物資訊學研究所 === 103 === Plant MIKC-type MADS-box transcription factors play an important role in controlling floral organ development. ABCDE model is an essential model to describe how angiosperm MADS-box genes regulate floral organ identity. Phylogenetic tree is the most common method for gene classification. However, in the previous study we find that when phylogenetic tree faces to massive, multi-species or incomplete sequences, it might lead to waste of time and error classification. NCB lab developed a web-based tool for angiosperm MADS-box gene classification by machine learning method, iMADS. However, the training dataset of the system was old, and didn’t treat with appropriate filtration. On the other hand, the five-class ABCDE model also cannot have a better description for the species which contain unique floral organ. In this study, we use phylogenetic analysis to group data by unsupervised clustering. The error classification data are modified by literatures. All of the genes which specifically express on the floral organ will be select as training dataset. The training model is constructed by two-stage. In addition to trying various features, we also extended five-class ABCDE model to eight-class, and constructed multiple prediction models, iMADS2.0 according to MADS-box gene domain characteristic by support vector machines. From the resulst, BLAST can get the best accuracy than other features of BindN and COILS. The datasets from the independent and the error classification of phylogenetic tree are submitted to prediction model for performance evaluation. The results showed that it could not only upgrade the prediction accuracy but correct every sequence to proper class. Finally, we used bioinformatics tools to discuss the relationship between physiochemical property of C-terminal domain and the regulation mechanism of transcription activation region. iMADS2.0 provides MADS-box gene predicted classification, other most similar predicted sequences and visualized expression patterns according to the region by user input. The web-based tool is freely available at http://predictor.nchu.edu.tw/iMADS2.