Models, Methods, and Criteria for Phylogeny Construction

这项探索性研究小额基金（SGER）由计算理论（TOC）项目、计算生物学活动（BIR）和系统生物学项目（DEB）共同资助，用于解决系统发育或进化树的构建问题。这个问题的形成和有效解决需要生物学家、统计学家和计算机科学家的多学科合作。解决这一问题的理想方法包括以下步骤：(a)观察现存物种的数据；(b)确定一种生物学模型（如Jukes-Cantor模型或Kimura双参数模型）；(c)基于步骤(b)的模型，设计目标函数和该函数的有效优化方法，使优化该目标函数的树为最适合该模型的树。不幸的是，这个理想的计划是不可能实现的，因为每一步都有障碍：(i)数据受到实验误差的影响，并且由于其在系统发育构建方法中的解释和使用而产生误差；似乎很难确定进化的精确生物学模式；（iii）给定进化的随机模型，优化树的一个候选是“最可能树”。考虑到什么才是真正的“最佳”模型的不确定性，一个模型下最可能的树在另一个稍有不同的模型下应该仍然是一个非常可能的树。证明这一点是一个非常困难的问题。本SGER提案的目标包括：(1)修改理想的方法，以消除困难(i) - (iii)；(2)利用分子生物学家提供的r-RNA和tufA序列数据测试各种模型；(3)根据实验结果，设计推断系统发育的一般方法；(4)将这些新模型与现有数据模型进行比较。此外，SGER奖的目标之一是启动宾夕法尼亚大学生物学家、统计学家和计算机科学家之间的多学科合作

This Small Grant for Exploratory Research (SGER), jointly funded by the Theory of Computing (TOC) Program, CCR, by Computational Biology Activity, BIR and by the Systemic Biology Program, DEB, addresses problems in the construction of phylogenies or evolutionary trees. The formulation and effective solution of this problem requires a collaborative multidisciplinary effort from biologists, statisticians and computer scientists. The ideal methodology for solving this problem would include the following steps:(a) Observe data on the species that exist today; (b) Identify a biological model (such as the Jukes-Cantor model or Kimura two parameter model); (c) Based on the model from Step (b), design an objective function and efficient optimization methods for this function so that the tree that optimizes this objective function is the tree that best fits the model. Unfortunately, this ideal program is impossible to realize, because of roadblocks at every step: (i) Data is subject to experimental error and to errors due to its interpretation and use in phylogeny construction methods;(ii) It seems difficult to identify a precise biological model for evolution; (iii) Given the stochastic model of evolution, one candidate for the optimizing tree is the ``most likely tree''. Given the uncertainty of what is actually the ``best'' model, a most likely tree under one model should still be a very likely tree under a slightly different model. Demonstrating this has proven to be a very difficult problem. The goals of this SGER proposal include:(1) Modification of the ideal methodology so that Difficulties(i) -- (iii) are removed; (2) Testing of various models using r-RNA and tufA sequence data supplied by a molecular biologist; (3) Using the experimental results, design of general methods for inferring phylogeny;(4) Comparison of these new models with existing models for the data. In addition, one of the goals of this SGER award is to initiate a multidisciplinary effort between the biologists, statisticians, and computer scientists at the University of Pennsylvania.***