Читать книгу Principles of Microbial Diversity - James W. Brown - Страница 80

Bayesian inference is a relatively new approach to tree construction. This approach starts with a random tree structure, random branch lengths, and random substitution parameters for an alignment, and the probability of the tree being generated from the alignment with these parameters is scored. Obviously the initial score is likely to be very poor. Then a random change is made in this tree (branch order, branch length, or substitution parameter) and the result is rescored. Then a choice is made whether to accept the change; this choice is partially random, but the greater the improvement in tree score, the more likely it is to be accepted. If the change is accepted, the process is repeated starting with this new tree; if the change is rejected, the process is repeated starting with the old tree. After many, many cycles of this process, the algorithm settles in to a collection of trees that are nearly optimal. Various tricks are used to keep the algorithm from getting stuck in local-scoring minimum zones.