Improvements and Extensions to the BLAST Algorithms

在PSI-BLAST搜索中偶尔出现的一个问题是 进化中的序列轮廓通过 在PSI-BLAST多个序列中包含非同源序列 对齐。在前几年，腐败现象一直很严重 通过改进PSI-BLAST统计来减少， 最重要的是考虑到非标准序列 组成。然而，最近观察到它在 有关PSI-BLAST配置文件可能损坏的文献 通过“同源过度扩展”，一个无法解决的问题 通过改进的统计数据进行补救。简而言之，这个问题 在其他情况下“真”对齐的边界 计算错误，导致对齐时间更长 应该是的。如果这样的对齐延伸到 在数据库中广泛出现的主题序列， 后续的PSI-BLAST迭代可以棘轮式的方式， 来包括完整的域，即使它确实包括了 在查询序列中不存在。问题不是由于 错误的统计数据，但错误的排列。一个解决方案 对于这个问题，在文献中已经提出了，但我们 已经采用了我们认为更好的补救措施，基于 将每一端的对齐修剪一定数量的位。 对这种方法的初步测试是有希望的，而且 这种方法的发展还在继续。没有任何出版物有 但结果却是。

A problem that occasionally arises in PSI-BLAST searches is the "corruption" of the evolving sequence profile through the inclusion of non-homologous sequences in the PSI-BLAST multiple alignment. In previous years, corruption has been greatly reduced through the improvement of PSI-BLAST statistics, most importantly by accounting for non-standard sequence composition. Recently, however, it has been observed in the literature that PSI-BLAST profiles may become corrupted through "homologous over-extension", a problem that can not be remedied by improved statistics. In brief, this problem arises when the boundaries of an otherwise "true" alignment are miscalculated, yielding the alignment longer than it should be. If such an alignment extends into a domain in the subject sequence that occurs widely in the database, subsequent PSI-BLAST iterations can, in a ratchet-like manner, come to include the complete domain, even though it does not exist in the query sequence. The problem is due not to faulty statistics, but to faulty alignment. One solution to this problem has been proposed in the literature, but we have adopted what we consider a better remedy, based upon trimming alignments at each end by a certain number of bits. Preliminary tests of this approach have been promising, and development of this method continues. No publications have yet resulted.