基因预测准确性的度量标准分析

An Analysis of the Measures for Gene Prediction Accuracy

分析了评价基因预测算法准确性的两个主要标准-相关系数CC(Correlation Coefficient)和近似相关系数AC(Approximate Correlation)的关系.首先在统一的概率框架下给出了CC和AC的统计描述,阐明了二者在概率意义上的差异,并系统的给出了|AC||CC|的证明以及等号成立的充要条件,最后用计算机模拟的方法分析了AC与CC之间大小差别的影响因素,得出预测准确性的高低和|FP-FN|的大小是两个影响|AC-CC|大小的主要原因.

The widely used measures AC and CC of gene prediction accuracy have been analyzed. First, statistical frameworks of AC and CC have been presented and their difference inessence have been explained. Then it is proved that ｜AC｜≥｜CC｜ holds except some particular cases and the conditions when AC = CC holds have been pointed out. Furthermore, It has been analyzed what influences the difference between AC and CC and concluded that both the prediction accuracy and the difference between FP and FN result in the difference between AC and CC.