基于神经网络集成方法预测膜蛋白类型

由Chou等人提出的预测膜蛋白分类的机器学习算法在近年来不断改进，使得预测膜蛋白类型的准确率越来越高。但是由于膜蛋白类分布不均衡而导致少数类的预测准确率非常低，使用神经网络集成方法能解决此问题。该方法中Bagging算法通过对多数类欠采样和少数类过采样来解决膜蛋白训练数据集不均衡问题。此外，用神经网络集成方法对已训练数据集和独立数据集进行分类测试，得出神经网络集成方法预测效果优于单个最好神经网络。该方法为解决蛋白质分类预测问题提供了一种新的策略，特别是训练数据集不均衡时，该方法的优势更加明显。

Prediction of Membrane Protein Types Based on Neural Network Ensemble Strategy

As a continuous effort to develop machine learning algorithms to predict membrane protein types that was initiated by Chou and Elrod, this study focuses on dealing with the problem of imbalanced training set of membrane protein types with the neural network ensemble. Bagging algorithm of the neural network ensemble has the advantage of dealing with imbalanced training set of membrane protein types by over-sampling minority classes and under-sampling majority classes. Furthermore, the perfor- mance of the neural network ensemble is found to be superior to the single best model from the results obtained through resubstitu-tion and independent dataset tests. The current approach represents a new strategy to deal with the problems of protein attribute pre-diction, especially when the training set is imbalanced, and hence is quite promising in the area of proteomies.