FASTA

FASTA is a DNA and proteinsequence alignment software package first described (as FASTP) by David J.
Lipman and William R. Pearson in 1985.[1] Its legacy is the FASTA format which is now ubiquitous in
bioinformatics.

History
The original FASTP program was designed for protein sequence similarity searching. FASTA
added the ability to do DNA:DNA searches, translated protein:DNA searches, and also provided
a more sophisticated shuffling program for evaluating statistical significance.[2] There are several
programs in this package that allow the alignment of protein sequences and DNA sequences.

Uses
FASTA is pronounced "fast A", and stands for "FAST-All", because it works with any alphabet,
an extension of "FAST-P" (protein) and "FAST-N" (nucleotide) alignment.

The current FASTA package contains programs for protein:protein, DNA:DNA,

protein:translated DNA (with frameshifts), and ordered or unordered peptide searches. Recent
versions of the FASTA package include special translated search algorithms that correctly handle
frameshift errors (which six-frame-translated searches do not handle very well) when comparing
nucleotide to protein sequence data.

In addition to rapid heuristic search methods, the FASTA package provides SSEARCH, an
implementation of the optimal Smith-Waterman algorithm.

A major focus of the package is the calculation of accurate similarity statistics, so that biologists
can judge whether an alignment is likely to have occurred by chance, or whether it can be used to
infer homology. The FASTA package is available from fasta.bioch.virginia.edu.

The web-interface to submit sequences for running a search of the European Bioinformatics
Institute (EBI)'s online databases is also available using the FASTA programs.

The FASTA file format used as input for this software is now largely used by other sequence
database search tools (such as BLAST) and sequence alignment programs (Clustal, T-Coffee,
etc.).

Search method
FASTA takes a given nucleotide or amino-acid sequence and searches a corresponding sequence
database by using local sequence alignment to find matches of similar database sequences.
The FASTA program follows a largely heuristic method which contributes to the high speed of
its execution. It initially observes the pattern of word hits, word-to-word matches of a given
length, and marks potential matches before performing a more time-consuming optimized search
using a Smith-Waterman type of algorithm.

The size taken for a word, given by the parameter ktup, controls the sensitivity and speed of the
program. Increasing the ktup value decreases number of background hits that are found. From
the word hits that are returned the program looks for segments that contain a cluster of nearby
hits. It then investigates these segments for a possible match.
Diagram from Book Protein Structure prediction - a practical approach from chapter Protein
Sequence Alignment and Database Scanning

There are some differences between fastn and fastp relating to the type of sequences used but
both use four steps and calculate three scores to describe and format the sequence similarity
results. These are:

 Identify regions of highest density in each sequence comparison. Taking a ktup to equal 1
or 2.

In this step all or a group of the identities between two sequences are found using a look
up table. The ktup value determines how many consecutive identities are required for a
match to be declared. Thus the lesser the ktup value: the more sensitive the search.
ktup=2 is frequently taken by users for protein sequences and ktup=4 or 6 for nucleotide
sequences. Short oligonucleotides are usually run with ktup = 1. The program then finds
all similar local regions, represented as diagonals of a certain length in a dot plot,
between the two sequences by counting ktup matches and penalizing for intervening
mismatches. This way, local regions of highest density matches in a diagonal are isolated
from background hits. For protein sequences BLOSUM50 values are used for scoring
ktup matches. This ensures that groups of identities with high similarity scores contribute
more to the local diagonal score than to identities with low similarity scores. Nucleotide
sequences use the identity matrix for the same purpose. The best 10 local regions selected
from all the diagonals put together are then saved.

 Rescan the regions taken using the scoring matrices. trimming the ends of the region to
include only those contributing to the highest score.

Rescan the 10 regions taken. This time use the relevant scoring matrix while rescoring to
allow runs of identities shorter than the ktup value. Also while rescoring conservative
replacements that contribute to the similarity score are taken. Though protein sequences
use the BLOSUM50 matrix, scoring matrices based on the minimum number of base
changes required for a specific replacement, on identities alone, or on an alternative
measure of similarity such as PAM, can also be used with the program. For each of the
diagonal regions rescanned this way, a subregion with the maximum score is identified.
The initial scores found in step1 are used to rank the library sequences. The highest score
is referred to as init1 score.

 In an alignment if several initial regions with scores greater than a CUTOFF value are
found, check whether the trimmed initial regions can be joined to form an approximate
alignment with gaps. Calculate a similarity score that is the sum of the joined regions
penalising for each gap 20 points. This initial similarity score (initn) is used to rank the
library sequences. The score of the single best initial region found in step 2 is reported
(init1).

Here the program calculates an optimal alignment of initial regions as a combination of

compatible regions with maximal score. This optimal alignment of initial regions can be
 rapidly calculated using a dynamic programming algorithm. The resulting score initn is
used to rank the library sequences.This joining process increases sensitivity but decreases
selectivity. A carefully calculated cut-off value is thus used to control where this step is
implemented, a value that is approximately one standard deviation above the average
score expected from unrelated sequences in the library. A 200-residue query sequence
with ktup2 uses a value 28.

 Use a banded Smith-Waterman algorithm to calculate an optimal score for alignment.

This step uses a banded Smith-Waterman algorithm to create an optimised score (opt) for
each alignment of query sequence to a database(library) sequence. It takes a band of 32
residues centered on the init1 region of step2 for calculating the optimal alignment. After
all sequences are searched the program plots the initial scores of each database sequence
in a histogram, and calculates the statistical significance of the "opt" score. For protein
sequences, the final alignment is produced using a full Smith-Waterman alignment. For
DNA sequences, a banded alignment is provided.