基于极限学习机的蛋白质相互作用预测

Predicting Protein-protein Interactions Based on Extreme Learning Machine Classifier

蛋白质间的相互作用参与到生命活动的全过程,对它的研究有助于揭示生命活动的机制.本文采用基于序列的计算方法对蛋白质间的相互作用进行预测.首先,将蛋白质序列转换为蕴含生物进化信息的位置特异性得分矩阵,然后抽取出数值化描述符的特征信息,最后利用极限学习机分类器对其潜在的相互作用进行预测.我们在酵母和幽门螺旋杆菌数据集上进行了实验,取得的准确率高达93.21%和84.87%.优异的实验结果证明,我们提出的模型可作为预测蛋白质相互作用的有效工具.

The interaction between proteins is involved in the whole process of life activity, and the study of it can help us reveal the mechanism of life activities. In this paper, a sequence based approach is used to predict protein - protein interactions. Firstly, the protein sequence is transformed into the position -specific score matrix which contains the biological evolution information, and then the feature information of the numerical descriptor is extracted. Finally, the potential interaction is predicted by the extreme learning machine classifier. We performed experiments on yeast and Helicobacter pylori data sets, with accuracy rates of up to 93.21% and 84.87%. The excellent experimental results show that our proposed model can he used as an effective tool for predicting protein - protein interactions.