What is Sequence Search?
BLAST Concepts

Why Does Sequence Search Exist?

Before the internet era, biologists had to manually compare sequences on paper — a completely impractical task once databases grew beyond a few hundred sequences. The question was: how do you efficiently find the needle (your query) in a haystack of millions of known sequences?

The naive approach — align your query against every single database entry using dynamic programming (the Smith-Waterman algorithm) — gives perfect results but is too slow for large databases. BLAST was invented in 1990 specifically to solve this speed problem while remaining accurate enough for practical biology.

Sequence search is used in bioinformatics for:

• Gene annotation — assigning function to newly sequenced genes by finding similar genes in model organisms
• Evolutionary analysis — finding orthologues (same gene, different species) and paralogues (duplicated genes)
• Primer design — checking that your PCR primers are specific and won't amplify off-target regions
• Contamination detection — checking whether your RNA-seq data contains sequences from an unexpected organism
• Metagenomics — identifying organisms in environmental samples by matching reads to known genomes

What is BLAST?

The word local distinguishes BLAST from global alignment tools like Needleman-Wunsch, which try to align two sequences from end to end. Local alignment is more useful for biology because:

• Genes often share only functional domains, not their entire sequence
• Proteins from distantly related organisms may share a conserved catalytic site but differ everywhere else
• Your query (a 200 bp fragment) is often much shorter than the database sequence (a 50 kb chromosome)

BLAST was created at the NCBI (National Center for Biotechnology Information) and is available both as a web tool at ncbi.nlm.nih.gov/blast and as a standalone command-line program you install locally. In this module, we focus on the command-line version — because that is what you use in automated pipelines.

BLAST Flavours — Which One Do You Use?

BLAST is not a single program — it is a family of programs, each designed for a specific combination of query type and database type. Choosing the wrong one is a common beginner mistake.

Key Concepts You Must Know

BLAST output contains a set of statistics that look intimidating at first but each has a precise biological meaning. You will see these in every paper, every pipeline, and every database result page.

E-value: A Deep Dive

The E-value depends on three things:

• The alignment score — higher scores give lower E-values
• The database size — a larger database gives a higher E-value for the same score (more chances for a random hit). This is why you cannot compare E-values from different databases directly
• The scoring matrix — BLOSUM62 for proteins, match/mismatch scores for DNA

The formula behind E-value is:

HSPs — High-Scoring Segment Pairs

BLAST output shows HSPs like this (pairwise format):

Score = 285 bits (308),  Expect = 2e-75
Identities = 147/158 (93%),  Gaps = 0/158 (0%)
Strand = Plus/Plus

Query  1    ATGGTCAAGCTCGTCGACGAGCTCATCACCAACGAGAAGGACCCCAACATCAAGCTGAC  60
            |||||||||||||||||||||||||||||| ||||||||||||||||||| ||||||||||
Sbjct  203  ATGGTCAAGCTCGTCGACGAGCTCATCATCAACGAGAAGGACCCCAACATCAAGCTGAC  262

Query  61   CGAGCAGCTCAAGAAGGAGATCGAGCGCATGCTGCACAAGGAGCCCGTCGACGTCACCA  120
            |||||||||||||||||| ||||||||||||||||||||||||||||||||||||||||||
Sbjct  263  CGAGCAGCTCAAGAAGGATATCGAGCGCATGCTGCACAAGGAGCCCGTCGACGTCACCA  322

Reading an HSP:

• Score — the raw + bit score pair. Bit score is in parentheses here; in tabular output it appears as a separate column
• Expect — the E-value for this HSP against this database
• Identities — matches / total aligned length. Pipes (|) show identical positions
• Gaps — number of gap characters inserted to improve the alignment
• Query / Sbjct lines — the actual aligned sequences with coordinates

How BLAST Works: The Algorithm in Plain English

BLAST achieves its speed through a clever heuristic — it avoids checking every possible alignment and instead uses a seed-and-extend strategy:

Exercises

# Use blastx
# Reason: your query is a nucleotide (transcript = DNA/RNA),
# but UniProt is a protein database.
# blastx translates your nucleotide query in all 6 reading frames
# and searches a protein database — perfect for gene annotation.

# blastn would be wrong: it searches a nucleotide DB
# blastp would be wrong: it requires a protein query

# Hit 1 (E = 3e-82) is far more reliable.
# An E-value of 3e-82 means you would expect this good an alignment
# to occur by chance only 3 × 10^-82 times — essentially never.
# This is very strong evidence of true homology.

# Hit 2 (E = 0.04) is borderline.
# You would expect ~0.04 false positives per search at this score.
# Standard threshold is E < 1e-5 for significance.
# Most researchers would NOT accept E = 0.04 as evidence of homology
# without additional evidence (e.g. conserved domain, phylogeny).

# Their E-value will be LARGER (less significant).
# E-value = K × m × n × e^(-λ × S)
# where n = effective database size.

# NCBI nr has ~20,000× more sequences than your local DB.
# So for the same alignment score S, the E-value scales proportionally:
# your E = 1e-8
# colleague's E ≈ 1e-8 × 20000 ≈ 2e-4

# This does NOT mean the alignment is worse — the underlying biology
# is the same. It just means there are more random hits in a bigger DB.
# Always report which database you searched when reporting E-values!

# This is an extremely high-confidence match.

# Score = 420 bits → very high alignment quality
# E = 5e-118 → essentially impossible to see by chance
# Identities = 94% → 198 out of 210 positions are identical
# Gaps = 1% → only 2 gap characters in 210 bp alignment

# Conclusion: this query and the database sequence are
# almost certainly the same gene or very recently diverged
# orthologues (e.g. same gene in closely related species).
# The 6% difference could be due to SNPs or allelic variation.