Question 12 of 20 Intermediate Concept
How does a short-read aligner such as BWA differ from BLAST in purpose and output?
Short answer
BLAST searches a database for any homologous sequence and reports ranked hits with E-values. A read aligner places millions of short, nearly identical reads at their single most likely position in one reference genome, and reports each placement with a mapping quality in SAM or BAM format.
Why
The two tools answer different questions. BLAST asks which sequences in a database are related to this query, including distant relatives, and reports statistical significance for each local alignment. A short-read aligner assumes the read came from the reference genome, with a few differences from sequencing errors or variants, and asks where it came from.
BWA, as described by Li and Durbin, indexes the reference with the Burrows-Wheeler transform so it can search for matches with a bounded number of mismatches and gaps very quickly, aligning short reads against a genome as large as the human one. Its output is the standard SAM format. Each alignment carries a mapping quality, the Phred-scaled probability that the placement is wrong, and pairs are used together, with Smith-Waterman alignment to rescue a read whose mate maps reliably.
Using the wrong tool causes trouble. BLAST is far too slow for a sequencing run and reports many hits per read. A read aligner tuned for near-identical matches will fail to place reads from a divergent species or a highly mutated region, and it gives no E-value for homology.
A strong answer covers
- BLAST finds homologs in a database, read aligners place reads in one genome
- Read aligners expect near-identical matches
- Burrows-Wheeler indexing makes genome-scale mapping fast
- Output is SAM or BAM with mapping quality, not E-values
- Read pairs help place and rescue reads
Common mistakes
- Using BLAST to map a sequencing run
- Using a short-read aligner to find distant homologs
- Treating mapping quality as a measure of homology
The interviewer might ask next
What would happen if you aligned reads from a related species to the human reference with default settings?