特征选择在蛋白质能量热点预测中的应用

蛋白质是细胞中的主要功能分子,是生命的物质基础,蛋白质的功能是通过蛋白质之间相互作用而发挥的,而蛋白质相互作用界面上只有很少数的被称之为＂能量热点＂的残基对相互作用贡献了大部分的结合自由能,如何识别这些能量热点是目前生物信息学领域比较热门的研究问题。其中基于机器学习的蛋白质能量热点识别中,特征选择方法的使用对识别模型的性能影响非常大。该文中,笔者通过对蛋白质能量热点识别中的特征选择方法的研究现状进行全面的分析,指出还存在的一些问题及以后改进的思路和方向,为蛋白质能量热点预测准确率的提高奠定基础。

基于递归特征消除法的蛋白质能量热点预测

基于蛋白质相互作用能量热点的特性,定义了残基接触数、溶剂可及性面积相对变化量所占比例等18个新特征。分别使用基于支持向量机(support vector machine,SVM)和基于F-Score的递归特征消除法进行特征选择,提出对应的预测模型SVM-RFE和F-Score-RFE用于蛋白质能量热点的预测。实验结果显示,在独立测试中F-Score-RFE模型的F1比当前预测性能最好的方法提高6.25%,表明所定义的新特征对蛋白质能量热点的识别具有较大的贡献。

Modeling and Analysis of Internal Heat Integrated Distillation Columns

A generalized steady-state model is being developed for an internal heat integrated distillation column (IHIDiC). A procedure incorporating the Newton-Raphson method is devised for solving the model equations. Separation of an ethanol-water binary mixture is simulated and analyzed with the model. Two pinch points are found within the process, making the separation extremely difficult and expensive. Two sharp fronts in the temperature and the composition profiles are being observed. With the introduction of heat integration, satisfactory separation may be obtained in a limited number of stages with lower reflux ratios. Increasing the pressure difference between the rectifying and the stripping sections, however, would bring about a reduced relative volatility between the two components involved, creating adverse separation performances. It is obvious that optimization of the IHIDiC is of prime importance.