July 9, 1986

	The programs FASTA, LFASTA and RDF2 are new versions of a "universal"
FASTP/FASTN program.  They are directly descended from FASTN, but instead
of using a fixed alphabet (ACGT or amino acids) and built-in scoring
matrices, all of the search parameters can be read in from a disk file.

	FASTA, TFASTA, LFASTA, RDF2, and the sequence analysis
programs AACOMP, GARNIER, (T)GREASE, CHOFAS, all read files in the
standard protein library format, i.e.

	>CODE - title line
	either protein sequence or DNA sequence


	>CODE2 - next sequence
	....

The FASTGB program reads the GENBANK floppy disk format for the DNA
sequence library.  It should only be used copies of these files.  You must
set FILES=16 (or greater) in a CONFIG.SYS file when using GFASTA, and you
should set the environment variable:

	set GBLIB=c:\bbnlib\

so the files can be found.

The scoring matrix file is determined by setting the environment variable
SMATRIX.  So by typing:

	set SMATRIX=c:\fasta\dna.mat

the program will use the DNA alphabet (A,C,G,T,U,R,Q,N, etc) and scoring
matrix used in FASTN.  If you do not set SMATRIX to anything, it uses an
internal alphabet and scoring matrix for proteins which is identical to
FASTP.  The configuration files on the disk are:

	codaa.mat	genetic code matrix for proteins
	idnaa.mat	identity matrix for proteins using PAM250 self scores
	iidnaa.mat	identity matrix for proteins using 1, 0
	prot.mat	pam250 matrix
	dna.mat	DNA alphabet and scoring matrix.
	altprot.mat	an experimental replacement for the PAM matrix
			developed by D. Lipman

The format of the SMATRIX file is:

line1:	;P or ;D, this comment (if present) is used to determine whether
		  amino acids (aa) or nucleotides (nt) should be used
		  int the program.
line2: scoring parameters
	KFACT BESTOFF BESTSCALE BKFACT BKTUP BESTMAX HISTSIZ

	KFACT is used in the "diagonal method" search for the best
initial regions, for proteins, KFACT = 4, for DNA, KFACT = 1.

	BESTOFF, BESTSCALE, BKFACT, BKTUP and BESTMAX are used to
calculate the cutoff score.  The bestcut parameter is calculated from
parameters 2 - 6. If N0 is the length of the query sequence:

	BESTCUT = BESTOFF + N0/BESTSCALE + BKFACT*(BKTUP-KTUP)
	if (BESTCUT>BESTMAX) BESTCUT=BESTMAX

HISTSIZ is the size of the histogram interval.

line3: deletion penalties.
	the first value is the penalty for the first residue in a gap,
the second value is the penalty charged to each subsequent residue in
a gap.

line4: end of sequence characters
	(these are not required, since IFASTA uses '>' for the
beginning of a sequence, but they are included).  If not used, the
line must be left blank.

line5: The alphabet

line6: the hash values for each letter in the alphabet.  This allows
several characters to be hashed to the same value, e.g. a DNA sequence
alphabet with A = adenosine, 1 = probably adenosine, P = purine, would have
each of these characters hash to 0.  The lowest hash value should be 0.

line7 - n:

	The lower triangle of the symmetric scoring matrix.  There should
be exactly as many lines as there are characters in the alphabet, and the
last line should have n-1 entries.  The program does not check for the
length of each line (perhaps it should), so it is easy to screw up a matrix
badly by having fewer entries in the scoring matrix than in the alphabet,
or vice-versa.

	In addition to the using the universal scoring matrix, FASTA has
several improvements from FASTN.  You can search libraries
that are made up of a number of files.  For example:

	FASTA test.seq @rodent.lib

would search the files named in the rodent.lib file.  If rodent.lib
contained:

	rat.lib
	mouse.lib
	hamster.lib

these three files would be searched by FASTA.  This can be used to search a
number of individual sequences without combining them into one file.

	FASTA also uses an improved method for calculating the initial
score, which allows the scores of several similar to be combined.  Thus
FASTA now reports three scores in the summary, 

	initn - the best score using multiple region alignment.
	init1 - the old fastp/n score from the best single region
	opt - 	an optimized score around the init0 region.  An optn score
		is not ready yet.