具有强化学习策略的决策树算法

传统决策树在对不平衡数据进行分类时,提高正类的权重和舍弃部分负类的信息,造成负类的预测精度较低。文章引入强化学习思想,提出一种基于马尔可夫决策过程的改进决策树方法。根据马尔可夫决策过程、当前分裂特征的标准化互信息和马修斯相关系数作为信息增益率的奖励或者惩罚,形成新的特征选择标准。实验结果表明,与其他传统方法相比,改进的马尔可夫决策树对非平衡数据整体的预测精度及负类预测精度均有提高。

Rebinding kinetics of dissociated amino acid ligand and carbon monoxide to ferrous microperoxidase-11 in aqueous solution

The rebinding kinetics of an amino acid ligand to ferrous microperoxidase-11 (MP11) after photolysis of aggregated ferrous MP11 was measured in aqueous solution with femtosecond transient visible absorption spectroscopy. The kinetics of CO rebinding to ferrous MP11 after photolysis of MP11CO was also measured in aqueous solution with femtosecond transient visible absorption spectroscopy. From these measurements, we found that either Val-11 or Lys-13 rebinds to ferrous MP11 exponentially with an 8 picosecond time constant in aggregated ferrous MP11 solution and that CO rebinds to ferrous MP11 nonexponentially with subnanosecond time scale in MP11CO solution. The kinetics of both the amino acid and CO rebinding to ferrous MP11 in MP11 system mimics that in carbon monoxide oxidation activator protein (CooA) or carboxymethyl cytochrome c (CmCytC) system. We also measured the kinetics of CO rebinding to ferrous MP11 in aqueous solution at different MP11CO concentrations and found that MP11CO concentration has an obvious effect on the kinetics of CO rebinding to ferrous MP11, where both the germinate yield and rate of CO rebinding to ferrous MP11 increase with the increase of MP11CO concentration. These findings suggested that the picosecond amino acid ligand rebinding process could disturb the proximal heme-ligand structure that possibly leads to the subnanosecond CO rebinding kinetics in MP11CO, CooACO and CmCytCCO systems.