Question 13 of 20 Intermediate Interpretation
How can low-complexity regions distort the interpretation of sequence similarity?
Short answer
Regions with biased composition, such as runs of a few amino acids or simple nucleotide repeats, can score highly against each other because of composition, not ancestry. That breaks the statistics behind E-values, so BLAST masks them with SEG for proteins and DUST for nucleotides by default.
Why
E-values assume the chance scores come from sequences with ordinary composition. Altschul's tutorial notes that as many as a quarter of protein residues sit in regions of highly biased composition. Two such regions can align with very high scores that owe almost nothing to the order of residues and everything to their composition.
The alignment is also biologically doubtful, because these regions probably arise by slippage, so a one-to-one correspondence of residues is not meaningful. NCBI's BLAST documentation gives examples: the protein stretch PPCDPPPPPKDKKKKDDGPP and the DNA stretch AAATAAAAAAAATAAAAAAT. BLAST filters proteins with SEG and nucleotides with DustMasker, and can also adjust scores for overall composition.
In practice, a hit whose alignment covers only a proline-rich or glutamine-rich segment, or a poly-A tail, should not be taken as homology. The filter can go too far. If a query is mostly low complexity, filtering can remove most of it and return no hits, and the BLAST FAQ suggests switching the filter off in that case, which then requires judging hits by hand.
A strong answer covers
- Biased composition inflates scores without shared ancestry
- Breaks the assumptions behind E-values
- SEG and DUST mask these regions by default
- Check whether a hit covers only a low-complexity segment
- Turning filters off needs manual judgment of each hit
Common mistakes
- Reporting homology from a match that spans only a repeat or tail
- Disabling the filter to get more hits and trusting the new E-values
- Assuming masked residues are deleted from the sequence
The interviewer might ask next
How would you compare two proteins that are mostly intrinsically disordered and low in complexity?