Sequence Alignment — Biological Sequence Alignment

Global alignment (Needleman–Wunsch) aligns the full length of both sequences from end to end, inserting gaps wherever necessary. It is appropriate when both sequences are of similar length and expected to be homologous along their entire extent (e.g., comparing orthologous full-length proteins). Local alignment (Smith–Waterman) finds the highest-scoring contiguous matching subsequences without requiring end-to-end coverage. It is preferred when looking for a conserved domain within a larger, otherwise divergent sequence, or when query and target lengths differ substantially.

BLAST is designed for database similarity search — you have one or a few sequences and want to find homologs in a large database. It returns scored hits ranked by E-value and handles sequences of arbitrary length. Short-read aligners (BWA, Bowtie2, HISAT2) are optimised for mapping millions of short reads (50–300 bp) generated by NGS platforms to a single reference genome with very high throughput. Use BLAST for annotation and homology searches; use a short-read aligner for read mapping in genomics, transcriptomics, or epigenomics experiments.

The E-value (expect value) in a BLAST search reports the number of alignments with an equal or better score that would be expected by chance in a database of the same size. An E-value of 0.001 means roughly one spurious hit per 1000 searches at this score level. Thresholds depend on context: 1e-5 is a common conservative cutoff for homology inference; 1e-3 may be used for remote homologs; values above 0.01 should be interpreted with caution. Always combine E-value with percent identity and alignment coverage — a low E-value with short alignment coverage may still be biologically uninformative.

Yes, but standard short-read aligners are not appropriate. Long reads from Oxford Nanopore or PacBio platforms are typically 1–50 kb and have higher per-base error rates than Illumina reads. Tools such as minimap2 and NGMLR are designed for long-read alignment; they use a different seed strategy (minimisers) and are tolerant of the elevated indel rates characteristic of long-read technologies.

The BLOSUM family is the standard choice for most protein alignment tasks. BLOSUM62 is the default in BLAST and performs well for sequences with roughly 30–60% identity. For more distantly related proteins (below 30% identity) use BLOSUM45 or BLOSUM30, which are calibrated on more divergent training sets and assign higher scores to rare but evolutionarily plausible substitutions. PAM matrices are an older alternative; PAM250 is roughly analogous to BLOSUM45 for distant comparisons.