Moreno-Hagelsieb Gabriel, Janga Sarath Chandra
Department of Biology, Wilfrid Laurier University, 75 University Avenue West, Waterloo, ON N2L 3C5, Canada.
Proteins. 2008 Feb 1;70(2):344-52. doi: 10.1002/prot.21564.
Phylogenetic profiles (PPs) are one of the most promising methods for predicting functional relationships by genomic context. The idea behind PPs is that if the products of two genes have a functional interdependence, the genes should both be either present or absent across genomes. One of the main problems with PPs is that evolutionarily close organisms tend to share a higher number of genes resulting in the overscoring of PP-relatedness. The proper measure of the overscoring effect of evolutionary redundancy requires examples of both functionally related genes (positive gold standards) and functionally unrelated genes (negative gold standards). Since experimentally verified functional interactions are only available for a few model organisms, there is a need for an alternative to gold standards. The presence of operons (polycistronic transcription units formed of functionally related genes) in prokaryotic genomes offers such an alternative. Genes in operons are located next to each other in the same DNA strand, and thus their presence should result in a higher proportion of predicted functional interactions among adjacent genes in the same strand than among adjacent genes in opposite strands. Under the preceding principle, we present a confidence value (CV) designed for evaluating predictions of functional interactions obtained using PPs. We first show that the CV corresponds to a positive predictive value calculated using experimentally known operons and further validate operon predictions based on this CV in other organisms using available microarray data. Then, we use a fixed CV of 0.90 as a reference to compare PP predictions obtained using different nonredundant genome datasets filtered at varying thresholds of genomic similarity. Our results demonstrate that nonredundant genome datasets increase the number of high-quality predictions by an average of 20%. Confidence values as those presented here should help compare other strategies and scoring systems to use phylogenetic profiles and other genomic context methods for predicting functional interactions.
系统发育谱（PPs）是通过基因组背景预测功能关系最有前景的方法之一。PPs背后的理念是，如果两个基因的产物具有功能上的相互依赖性，那么这些基因在整个基因组中应该要么都存在，要么都不存在。PPs的一个主要问题是，进化上相近的生物体往往共享更多的基因，导致PP相关性得分过高。要恰当衡量进化冗余的过高得分效应，需要功能相关基因（正金标准）和功能不相关基因（负金标准）的例子。由于只有少数模式生物有经过实验验证的功能相互作用，因此需要一种替代金标准的方法。原核生物基因组中操纵子（由功能相关基因组成的多顺反子转录单元）的存在提供了这样一种替代方法。操纵子中的基因在同一条DNA链上彼此相邻，因此它们的存在应该会导致同一条链上相邻基因之间预测的功能相互作用比例高于相反链上相邻基因之间的比例。基于上述原则，我们提出了一个置信值（CV），用于评估使用PPs获得的功能相互作用预测。我们首先表明，CV对应于使用实验已知操纵子计算出的阳性预测值，并使用可用的微阵列数据在其他生物体中基于此CV进一步验证操纵子预测。然后，我们使用固定的CV值0.90作为参考，比较使用在不同基因组相似性阈值下过滤的不同非冗余基因组数据集获得的PP预测。我们的结果表明，非冗余基因组数据集将高质量预测的数量平均增加了20%。本文提出的置信值应有助于比较其他策略和评分系统，以便使用系统发育谱和其他基因组背景方法来预测功能相互作用。
