Distance and Likelihood Trees, and What Branch Support Means

An ordinary sequence bootstrap reuses an observed alignment. To make one replicate, choose columns with replacement until there are as many columns as in the original. Keep each selected column intact across all taxa. Repeated selections copy a column; unselected columns are absent from that replicate. Apply the same tree-building procedure to each replicate. Count how often a particular split occurs in the resulting trees. That proportion is its ordinary bootstrap support. The procedure explores sensitivity to resampling the observed columns. It does not collect new sequences. Running more replicates can make the estimated recovery frequency more stable, but cannot replace missing biological evidence. Read the reported method before interpreting a support value: other statistics are calculated differently.

At an aligned position, two matching bases describe the present states. They do not give a complete history. A base could have changed and later returned to its starting state, leaving no difference between the endpoints. Raw distances summarize observed dissimilarity. Model-corrected distances estimate accumulated change under substitution assumptions. Neither kind is an alignment gap: a gap is a feature of how sequences are aligned, and its treatment must be stated separately. A distance method fits a tree to pairwise estimates. Maximum likelihood evaluates the observed alignment under a candidate tree and model. Both approaches involve assumptions. A disagreement between their trees is a reason to examine the evidence and assumptions, not proof that either method recovered the historical relationships.

A weakly supported split can have several causes. There may be too few informative sites to distinguish the candidate relationships. Different sites may favor conflicting arrangements. Alignment problems or unsuitable model assumptions may also affect the result. Start with the limitation suggested by the evidence. For a short, well-checked alignment, additional informative sites may help. If alignment quality is doubtful, examine whether the compared positions are homologous. Assess the usefulness of taxon sampling and compare plausible models where their assumptions are a concern. These checks do not promise a higher support value. They help determine what the evidence can support. Report unresolved relationships openly. Even high ordinary bootstrap recovery can repeat a systematic error, so support values do not make model checking unnecessary.