School of Computer Science, China University of Geosciences, Wuhan, China

---

Abstract—Multiple Sequence alignments are a critical step in phylogeny inference. There is a lack of an appropriate approach which is capable of 1) finding the best global alignment and 2) automating and reproducing manual editing. Progressive alignment is an effective method for multiple Sequence alignments. However, its application in practice has also long been largely hampered because the alignment regions are not homologous to maximize the alignment score. The standard practice in phylogenetics involves manual editing of alignments and manual editing is a non-trivial task. Aiming at these problems, this study 1) uses SVM to capture the neighborhood of a site to automate and reproduce manual editing, and 2) builds the procedure of SVM Model Training and Automatic Annotation. Experimental results demonstrate that a SVM-based classifier can reproduce the manual editing tasks with an accuracy of 95.5%. This method is stable to both RBF parameters (Gamma and C) and clearly outperforms GBLOCKS and AL2CO, which are conventional editing/annotating methods. The classification accuracy achieved by the proposed method is always much higher than those achieved by the counterpart methods.

Index Terms—Multiple Sequence Alignments, machine learning, automatic annotation

[PDF]

Cite: Jiansi Ren, "SVM-based Automatic Annotation of Multiple Sequence Alignments," Journal of Computers vol. 9, no. 5, pp. 1109-1116, 2014.