Question 14 of 20 Intermediate Scenario
A homolog is missed by BLASTP but found by a profile search. Why might that happen?
Short answer
A single-sequence search scores every position with the same general matrix, so a distant relative that keeps only the family's key residues scores poorly. A profile from a multiple alignment weights each position by how conserved it is across the family, which makes those faint signals visible.
Why
BLASTP compares one query with one target using a matrix like BLOSUM62 that treats every position alike. In a distant relative, most positions have drifted, and only some, such as catalytic residues or structural core, stay conserved. A profile captures that.
PSI-BLAST builds a position-specific score matrix from significant hits and searches again, and its authors report that it is in many cases much more sensitive to weak but biologically relevant similarities. Profile hidden Markov models, as in HMMER3, also model position-specific insertion and deletion probabilities. Eddy reports HMMER3 as substantially more sensitive than its predecessor and about as fast as BLAST, with benchmarks built from true homologs no more than 25 percent identical to the query family.
Pearson adds that family-based searches can identify far more homologs in a single search at little extra cost. Profiles bring their own risk. If an unrelated sequence enters an iteration, the profile drifts and pulls in more unrelated sequences, and Pearson notes that alignment overextension into neighboring non-homologous regions is a common cause of errors in PSI-BLAST.
A strong answer covers
- Single-sequence matrices treat all positions alike
- Profiles weight conserved positions more strongly
- PSI-BLAST builds iterative position-specific matrices
- Profile HMMs also model position-specific gaps
- Iterative searches can drift if a false hit enters the profile
Common mistakes
- Concluding no homolog exists after one BLASTP search
- Accepting every new hit from later PSI-BLAST iterations without checking
- Building a profile from an unchecked, poorly aligned set of sequences
The interviewer might ask next
How would you confirm that a sequence added in a late PSI-BLAST iteration is a true family member?