Markov model plus k-word distributions: a synergy that produces novel statistical measures for sequence comparison
Markov model plus k-word distributions: a synergy that produces novel statistical measures for sequence comparison
复制标题
马尔可夫模型加上 k 字分布：产生用于序列比较的新颖统计测量的协同作用
DOI：
10.1093/bioinformatics/btn436
复制
发表时间：
2008-10-15
期刊：
影响因子：
5.8
通讯作者：
Wang, Tianming
中科院分区：
文献类型：
--
作者：
Dai, Qi;Yang, Yanchun;Wang, Tianming
动机
许多提出的统计方法可以有效地比较生物序列，以进一步推断它们的结构，功能和进化信息。它们在精神上是相关的，因为所有用于序列比较的想法都试图使用关于k字分布、马尔可夫模型或两者的信息。通过将k字分布直接加入到马尔可夫模型中，我们研究了两种新的序列比较统计度量，称为wre.k.r和S2.k.r。
结果
通过相似性检索、功能相关调控序列评价和系统发育分析对所提方法进行了验证。这为我们的措施提供了系统和定量的实验评估。此外，我们将我们的成就与基于对齐或无对齐的成就进行了比较。我们把实验分成两组。第一个，通过ROC（受试者工作曲线）分析，旨在评估我们的统计措施的内在能力，从数据库中搜索相似的序列，并区分功能相关的调控序列无关的序列。第二个目的是评估我们的统计措施是如何用于系统发育分析。实验评估表明，我们的相似性措施，旨在将k字分布到马尔可夫模型是更有效的。
MOTIVATION
Many proposed statistical measures can efficiently compare biological sequences to further infer their structures, functions and evolutionary information. They are related in spirit because all the ideas for sequence comparison try to use the information on the k-word distributions, Markov model or both. Motivated by adding k-word distributions to Markov model directly, we investigated two novel statistical measures for sequence comparison, called wre.k.r and S2.k.r.
RESULTS
The proposed measures were tested by similarity search, evaluation on functionally related regulatory sequences and phylogenetic analysis. This offers the systematic and quantitative experimental assessment of our measures. Moreover, we compared our achievements with these based on alignment or alignment-free. We grouped our experiments into two sets. The first one, performed via ROC (receiver operating curve) analysis, aims at assessing the intrinsic ability of our statistical measures to search for similar sequences from a database and discriminate functionally related regulatory sequences from unrelated sequences. The second one aims at assessing how well our statistical measure is used for phylogenetic analysis. The experimental assessment demonstrates that our similarity measures intending to incorporate k-word distributions into Markov model are more efficient.