| Summary: | 碩士 === 國立中興大學 === 生物科技學研究所 === 105 === Recently, most SUMOylation prediction tools are using algorism, protein physicochemical and biochemical properties or consensus motif to predict modification sites. But these tools rarely mention the effect of other post translational modification (PTM) on sumoylation prediction. In this study, we developed a sumoylation prediction system based on machine learning approach employing SVM (support vector machine) and also updated sumoylation consensus motif and related information. In the feature coding, we encoded binary code and protein properties based on amino acid sequence. Besides, we encoded other PTM distribution as functional feature and secondary information as structure feature. We tested the prediction system that removed the post-modification distribution code and found that the prediction system had lower accuracy than the non-removed post-modification coding. Top fifty percent of feature rankings from the two feature selection methods, eighty and forty percent of all post-modification distributions were included. Those result show the influence of other post-modification sites in this study. In addition, we analyzed the number of the post-modification distributions under the central lysine and window size 21 rules, and we provided some of our findings and recommended post-modification types that could be considered. Finally, this study developed a new sumoylation prediction tool called SUMOdig. We tested SUMOdig composed of positive and randomly negative in ratio 1:1 at twenty times. The sumoylation sites with an average Matthew’s correlation coefficient is equal to 0.5114.
|
|---|
