利用伪氨基酸组分和支持向量机预测抗冻蛋白

Predicting antifreeze proteins by using pseudo amino acid composition and support vector machine

抗冻蛋白是一类具有提高生物抗冻能力的蛋白质。抗冻蛋白能够特异性的与冰晶相结合,进而阻止体液内冰核的形成与生长。因此,对抗冻蛋白的生物信息学研究对生物工程发展,提高作物抗冻性有重要的推动作用。本文采用由400条抗冻蛋白序列和400条非抗冻蛋白序列构成数据集,以伪氨基酸组分为特征,利用支持向量机分类算法预测抗冻蛋白,对训练集预测精度达到91.3%,对测试集预测精度达到78.8%。该结果证明伪氨基酸组分能够很好的反映抗冻蛋白特性,并能够用于预测抗冻蛋白。

Antifreeze protein （AFP） is a kind of protein that can improve the antifreeze capability of organisms. They specifically bind to ice crystals to inhibit growth and recrystallization of ice. It is very important for bioengi- neering and for improving antifreeze capability of crop to accurately identify AFPs. The present study constructed a benchmark dataset including 400 AFPs and 400 non-AFPs. By using pseudo amino acid composition as parameters, support vector machine was applied to perform prediction. We finally achieved overall accuracies of 91.3% and 78.8%, respectively for training set and test set. These results suggest that pseudo amino acid composition can describe the characteristics of AFPs and can be used for AFPs prediction.