Multiple sequence alignment tool

How does the built-in aligner work?

It is a progressive aligner, the same three-stage shape the standard tools use. First, every pair of sequences is compared by shared k-mer counts to give a fast distance estimate without aligning anything. Those distances build a UPGMA guide tree, which fixes the order in which sequences are merged. Sequences are then aligned into profiles along that tree, most similar first, with a profile-to-profile dynamic programme that uses affine gap penalties: a large cost to open a gap and a small cost to extend it, so indels come out as a few long gaps rather than many short ones. Protein columns are scored with BLOSUM62. The whole recurrence is written against NumPy arrays, which is why a set of plasmid-sized sequences aligns in a fraction of a second rather than needing a queued job.

When should I use MAFFT, MUSCLE or Clustal Omega instead?

The built-in aligner is the right default for the cases a cloning lab meets most: a handful of closely related constructs, a gene family, a set of Sanger reads or homologous proteins, and it is always available. The dedicated aligners have iterative refinement stages the progressive method lacks, which matters for large, divergent sets or when the alignment feeds a phylogenetic analysis. When any of them is installed on the server it appears in the algorithm picker; the picker only offers what that deployment can actually run.

• MAFFT: fast and accurate on large sets; the usual choice for hundreds of sequences.
• MUSCLE: strong on protein families; iterative refinement of the progressive alignment.
• Clustal Omega: scales to very large sets using HMM profiles.

Reading the result

The alignment wraps into blocks that fit the window, so a multi-kilobase alignment reads top to bottom rather than scrolling sideways. Under each block a conservation track shows how uniform the column is, and a consensus row gives the majority residue. Differences-only mode hides every column where all sequences agree, which turns a screen of identical bases into the handful of positions that matter. Clicking a column lists what each sequence has there. The identity matrix reports pairwise identity across the finished alignment as a heat map, and the tree beside it is UPGMA over that same matrix, so the tree can never disagree with the numbers next to it.

What the tree is, and is not

The dendrogram is a UPGMA clustering of pairwise identity from the alignment. It is a faithful summary of how similar the sequences are to one another and a reasonable first look at a gene family, but it is not a phylogenetic inference: there is no substitution model, no neighbour-joining or maximum likelihood, and no bootstrap support. For a publication phylogeny, export the alignment as aligned FASTA and build the tree in a dedicated tool.