As   sequences diverge over time, the observed number of changes underestimates   the real number of changes. Imagine a nucleotide that begins as A. In one   lineage it is replaced by a C and then again by a T, the two changes would   still only cause one difference in the alignment. Also, sites that match in   the alignment could have changed independently, but ended up the same.   Fortunately, the extent of real divergence can be estimated from the total   observed amount of divergence. These ‘multi-hit correction’methods also differ   in their sophistication. However, none can work miracles: as the number of changes   increases, the amount of information from the alignment decreases and we   approach saturation, in which case the data are useless.

随着时间的推移，序列发生了变化，观察到的（碱基）变化数量低估了变化的实际数量。想象一个位点开始核苷酸为A。在一个谱系中，它被C替换，然后又被T替换，这两个变化导致比对仍然只会观察到一个差异。此外，比对位点可能会独立改变，但最终会相同。幸运的是，实际分化程度的范围可以通过观测到的总分化程度来估计。这些“多次命中校正”方法的复杂性也不同。然而，没有一个能创造奇迹：随着变化的数量增加，来自比对的信息量减少，我们就接近饱和，在这种情况下，数据是无用的。