Question 10 of 20 Fundamentals Interpretation
How does a BLAST bit score differ from the raw alignment score?
Short answer
The raw score depends on the matrix and gap penalties used, so it has no fixed meaning. The bit score rescales it with the scoring system's statistical parameters into standard units, so it can be compared across searches, and only the search space is then needed to get an E-value.
Why
A raw score is the sum of substitution scores minus gap penalties. Its size depends entirely on the scoring system, so citing a raw score alone is, in Altschul's words, like citing a distance without saying feet, meters or light years. BLAST statistics describe chance scores with two parameters, lambda and K, specific to the scoring system.
Normalizing the raw score with them gives the bit score, which carries the statistics of the scoring system inside it. The E-value then depends only on the bit score and the size of the search space, roughly the query length times the database length. Pearson draws two practical rules from this.
Each extra 10 bits makes a match about 1,000 times more significant, because 2 to the power of 10 is 1,024. And for average-length proteins a bit score of 50 is almost always significant, while 40 bits is significant only in small databases. That makes the bit score a steadier guide than the E-value when database sizes differ.
Bit scores also scale with alignment length, so a long weak alignment and a short strong one can reach the same score.
A strong answer covers
- Raw scores depend on the scoring system
- Bit scores normalize with lambda and K
- E-value needs only bit score and search space
- Ten extra bits is about a thousandfold more significance
- Around 50 bits is a practical significance rule for proteins
Common mistakes
- Comparing raw scores between searches run with different matrices
- Treating the bit score as a percent identity
- Ignoring that bit scores grow with alignment length
The interviewer might ask next
Why does the same alignment get a different E-value when the database grows but keep its bit score?