thebench.bio #1 Ecosystem for the Bio-Tech Industry
All Sequence Alignment questions

Question 16 of 20 Intermediate Practical

How would you build and check a multiple sequence alignment for a protein family?

Short answer

Collect true homologs, remove fragments and near duplicates, align with a progressive or consistency-based tool such as Clustal Omega or MAFFT, and inspect the result. Check that known conserved residues line up, then trim unreliable columns before using it for phylogeny or profiles.

Why

Most multiple aligners use the progressive approach. They build a guide tree and align sequences into larger and larger groups following its branching order. The Clustal Omega paper notes this is a greedy algorithm: a mistake made early is never corrected later.

It also describes how Clustal Omega scales to very large families by building guide trees with a fast embedding method, and how an external profile HMM, for example from Pfam, can help align new sequences. Because errors propagate, input quality matters. Fragments, unrelated sequences and very divergent members distort the guide tree and the alignment.

After aligning, check known biology: catalytic residues, cysteines in disulfides or a well-known motif should form clean columns. Many downstream analyses also benefit from trimming. The trimAl authors note that removing poorly aligned regions improves later analyses and offer automated modes such as gappyout.

Trimming should be done for a purpose. A phylogeny wants reliable columns, but a study of insertions in one lineage would lose exactly what it needs.

How to work it through

  1. Gather candidate sequences with a sensitive search and confirm they are homologs.
  2. Remove fragments, obvious misassemblies and near-identical duplicates.
  3. Align with a tool suited to the family size, adding a profile from a curated family when one exists.
  4. Inspect the alignment and confirm that known conserved residues and motifs line up.
  5. Trim poorly aligned or gap-rich columns with a tool such as trimAl when the next step needs reliable columns.
  6. Record the tool versions, parameters and sequences removed.

An example

A team aligns 420 kinase domain sequences. A first alignment places the catalytic lysine in three different columns. Removing 35 fragments shorter than half the domain and 12 sequences from a misannotated gene model, then realigning with a kinase profile, puts the lysine and the DFG motif in single clean columns. trimAl gappyout removes 18 percent of columns, mostly in a variable loop, before tree building.

Where the answer stops holding

No aligner is reliable for very divergent sequences or long variable loops, so some columns remain guesses. Benchmarks based on structural reference alignments favor structured regions. Trimming can remove phylogenetic signal as well as noise.

A strong answer covers

  • Clean the input: true homologs, no fragments or duplicates
  • Progressive alignment errors are never corrected later
  • Use external profiles when available
  • Check that known conserved residues align
  • Trim with a purpose and record what was removed

Common mistakes

  • Feeding every BLAST hit into the aligner without filtering
  • Trusting an alignment without looking at known motifs
  • Trimming aggressively before an analysis that needs variable regions