 1997 Oxford University Press Nucleic Acids Research, 1997, Vol. 25, No.

17 3389–3402

Gapped BLAST and PSI-BLAST: a new generation of

protein database search programs
Stephen F. Altschul*, Thomas L. Madden, Alejandro A. Schäffer1, Jinghui Zhang,
Zheng Zhang2, Webb Miller2 and David J. Lipman

National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health,
Bethesda, MD 20894, USA, 1Laboratory of Genetic Disease Research, National Human Genome Research
Institute, National Institutes of Health, Bethesda, MD 20892, USA and 2Department of Computer Science and
Engineering, Pennsylvania State University, University Park, PA 16802, USA

Received June 20, 1997; Revised and Accepted July 16, 1997

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
ABSTRACT BLAST is a heuristic that attempts to optimize a specific
similarity measure. It permits a tradeoff between speed and
The BLAST programs are widely used tools for sensitivity, with the setting of a ‘threshold’ parameter, T. A higher
searching protein and DNA databases for sequence value of T yields greater speed, but also an increased probability
similarities. For protein comparisons, a variety of of missing weak similarities. The BLAST program requires time
definitional, algorithmic and statistical refinements proportional to the product of the lengths of the query sequence
described here permits the execution time of the and the database searched. Since the rate of change in database
BLAST programs to be decreased substantially while sizes currently exceeds that of processor speeds, computers
enhancing their sensitivity to weak similarities. A new running BLAST are subjected to increasing load. However, the
criterion for triggering the extension of word hits, conjunction of several new algorithmic ideas allow a new version
combined with a new heuristic for generating gapped of BLAST to achieve improved sensitivity at substantially
alignments, yields a gapped BLAST program that runs augmented speed. This paper describes three major refinements
at approximately three times the speed of the original. to BLAST.
In addition, a method is introduced for automatically (i) For increased speed, the criterion for extending word pairs
combining statistically significant alignments pro- has been modified. The original BLAST program seeks short
duced by BLAST into a position-specific score matrix, word pairs whose aligned score is at least T. Each such ‘hit’ is then
and searching the database using this matrix. The extended, to test whether it is contained within a high-scoring
resulting Position-Specific Iterated BLAST (PSI- alignment. For the default T value, this extension step consumes
BLAST) program runs at approximately the same most of the processing time. The new ‘two-hit’ method requires
speed per iteration as gapped BLAST, but in many the existence of two non-overlapping word pairs on the same
cases is much more sensitive to weak but biologically diagonal, and within a distance A of one another, before an
relevant sequence similarities. PSI-BLAST is used to extension is invoked. To achieve comparable sensitivity, the
uncover several new and interesting members of the threshold parameter T must be lowered, yielding more hits than
BRCT superfamily. previously. However, because only a small fraction of these hits
are extended, the average amount of computation required
INTRODUCTION decreases.
(ii) The ability to generate gapped alignments has been added.
Variations of the BLAST algorithm (1) have been incorporated The original BLAST program often finds several alignments
into several popular programs for searching protein and DNA involving a single database sequence which, when considered
databases for sequence similarities. BLAST programs have been together, are statistically significant. Overlooking any one of
written to compare protein or DNA queries with protein or DNA these alignments can compromise the combined result. By
databases in any combination, with DNA sequences often introducing an algorithm for generating gapped alignments, it
undergoing conceptual translation before any comparison is becomes necessary to find only one rather than all the ungapped
performed. We will use the blastp program, which compares alignments subsumed in a significant result. This allows the T
protein queries to protein databases, as a prototype for BLAST, parameter to be raised, increasing the speed of the initial database
although the ideas presented extend immediately to other scan. The new gapped alignment algorithm uses dynamic
versions that involve the translation of a DNA query or database. programming to extend a central pair of aligned residues in both
Some of the refinements described are applicable as well to directions. For speed, earlier heuristic methods (2,3) confined the
DNA–DNA comparison, but have yet to be implemented. alignments produced to a predefined strip of the dynamic

* To whom correspondence should be addressed. Tel: +1 301 496 2475; Fax: +1 301 480 9241; Email: altschul@ncbi.nlm.nih.gov
3390 Nucleic Acids Research, 1997, Vol. 25, No. 17

programming path graph (4). Our approach considers only where N = mn is the search space size (8–10). If a protein is
alignments that drop in score no more than Xg below the best compared to a whole database rather than a single sequence, n is
score yet seen. The algorithm is able thereby to adapt the region the database length in residues. Equation 2 may be inverted to
of the path graph it explores to the data. yield S′ = log2(N/E), the normalized score required to achieve a
(iii) BLAST searches may be iterated, with a position-specific particular E-value. In a typical current database search, a protein
score matrix generated from significant alignments found in of length 250 might be compared to a protein database of 50 000 000
round i used for round i + 1. Motif or profile search methods total residues. To achieve a marginally significant E-value of
frequently are much more sensitive than pairwise comparison 0.05, a normalized score of ∼38 bits is necessary.
methods at detecting distant relationships. However, creating a While the theory just outlined has not been proved for gapped
set of motifs or a profile that describes a protein family, and local alignments and their associated scores, computational
searching a database with them, typically has involved running experiments strongly suggest that it remains valid (3,12–15). The
several different programs, with substantial user intervention at statistical parameters λ and K, however, are no longer supplied by
various stages. The BLAST algorithm is easily generalized to use theory but must be estimated using comparisons of either
an arbitrary position-specific score matrix in place of a query simulated or real but unrelated sequences. To distinguish below
sequence and associated substitution matrix. Accordingly, we whether a given set of parameters λ and K refer to gapped or
have automated the procedure of generating such a matrix from ungapped alignments, we use the subscripts g and u respectively.

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
the output produced by a BLAST search, and adapted the BLAST When gaps are not allowed, a further important theorem states
algorithm to take this matrix as input. The resulting Position- that within HSPs the aligned pair of letters (i,j) tends to occur with
Specific Iterated BLAST, or PSI-BLAST, program may not be as the ‘target frequency’:
sensitive as the best available motif search programs, but its speed
and ease of operation can bring the power of these methods into q ij  PiPje
usij 3
more common use.
After describing these refinements to BLAST in greater detail, The qij of equation 3 sum to 1; indeed, λu is calculated to be the
we consider several biological examples for which the sensitivity unique positive number for which this is the case (8,9). The scores
and speed of the program are greatly enhanced. sij are optimal for detecting alignments with these particular target
frequencies (8,10), and by inverting equation 3 to sij =
[ln(qij/PiPj)]/λu, scores may be chosen, with arbitrary scale, that
STATISTICAL PRELIMINARIES correspond to any desired set of qij. The popular PAM (16,17) and
BLOSUM (18) substitution matrices are constructed with explicit
To analyze the BLAST algorithm and its refinements, we need use of this log-odds formula. No corresponding result has been
first to review the statistics of high-scoring local alignments. established for gapped alignment scoring systems. However, if
BLAST employs a substitution matrix, which specifies a score sij the gap costs used are sufficiently large, it is expected that the
for aligning each pair of amino acids i and j. Given two sequences target frequencies observed in high-scoring local alignments of
to compare, the original BLAST program seeks equal-length random sequences will not differ greatly from those for the
segments within each that, when aligned to one another without no-gap case.
gaps, have maximal aggregate score. Not only the single best
segment pair may be found, but also other locally optimal pairs REFINEMENT OF THE BASIC ALGORITHM: THE
(3,5–7), whose scores cannot be improved by extension or TWO-HIT METHOD
trimming. Such locally optimal alignments are called ‘high-scor-
ing segment pairs’ or HSPs. The central idea of the BLAST algorithm is that a statistically
For the sake of the statistical theory, we assume a simple protein significant alignment is likely to contain a high-scoring pair of
model in which the amino acids occur randomly at all positions aligned words. BLAST first scans the database for words
with background probabilities Pi. We require that the expected (typically of length three for proteins) that score at least T when
score for two random amino acids PPs i j ij be negative. Given
aligned with some word within the query sequence. Any aligned
word pair satisfying this condition is called a hit. The second step
i,j
the Pi and sij, the basic theory (8,9) yields two calculable of the algorithm checks whether each hit lies within an alignment
parameters, λ and K, which can be used to convert nominal HSP with score sufficient to be reported. This is done by extending a
scores to normalized scores, thereby rendering all scoring hit in both directions, until the running alignment’s score has
systems directly comparable from a statistical perspective. The dropped more than X below the maximum score yet attained. This
normalized score S′ for an HSP is given by the equation: extension step is computationally quite costly; with the T and X
parameters necessary to attain reasonable sensitivity to weak
alignments, the extension step typically accounts for >90% of
S 
S
ln K 1 BLAST’s execution time. It is therefore desirable to reduce the
ln 2
number of extensions performed.
In this paper, a nominal score is given without units, while a score Our refined algorithm is based upon the observation that an
normalized by equation 1 is said to be expressed in bits (10,11). HSP of interest is much longer than a single word pair, and may
When two random protein sequences of sufficient lengths m and therefore entail multiple hits on the same diagonal and within a
n are compared, the number E of distinct HSPs with normalized relatively short distance of one another. (The diagonal of a hit
score at least S′ expected to occur by chance is well approximated involving words starting at positions (x1, x2) of the database and
by: query sequences may be defined as x1 – x2. The distance between
two hits on the same diagonal is the difference between their first
E  N2 S 2 coordinates.) This signature may be used to locate HSPs more
 3391

Nucleic Acids
Nucleic Acids Research,
Research,1994,
1997,Vol.
Vol.22,
25,No.
No.117 3391

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
Figure 2. The BLAST comparison of broad bean leghemoglobin I (87)
Figure 1. The sensitivity of the two-hit and one-hit heuristics as a function of (SWISS-PROT accession no. P02232) and horse β-globin (88) (SWISS-PROT
HSP score. Using the BLOSUM-62 amino acid substitution matrix (18), and the accession no. P02062). The 15 hits with score at least 13 are indicated by plus
target frequencies qij implied by equation 3 and the background amino acid signs. An additional 22 non-overlapping hits with score at least 11 are indicated
frequencies Pi of Robinson and Robinson (20), 100 000 model HSPs were by dots. Of these 37 hits, only the two indicated pairs are on the same diagonal
generated for each of the nominal scores 37–92, corresponding to normalized and within distance 40 of one another. Thus the two-hit heuristic with T = 11
scores 19.9–45.1 bits. It was determined by inspection whether each HSP failed triggers two extensions, in place of the 15 extensions invoked by the one-hit
to contain two non-overlapping length-3 word pairs with nominal score at least heuristic with T = 13. Because this is just one example, the relative numbers of
11, and within a distance 40 of one another, or a single length-3 word pair with hits and extensions at the various settings of T correspond only roughly to the
nominal score at least 13. The corresponding probabilities of missing an HSP ratios found in a full database search. An ungapped extension of the leftward
using the two-hit heuristic with T = 11, and the one-hit heuristic with T = 13, of the two hit pairs yields an HSP with nominal score 45, or 23.6 bits, calculated
are plotted as a function of normalized HSP score. The two-hit method is more using λu and Ku.
sensitive for HSPs with score at least 33 bits.

we evaluate the sensitivity of the one-hit and two-hit BLAST

efficiently. Specifically, we choose a window length A, and heuristics using these HSPs.
invoke an extension only when two non-overlapping hits are The one-hit method will detect an HSP if it somewhere contains
found within distance A of one another on the same diagonal. Any a length-W word of score at least T. For W = 3 and T = 13, Figure
hit that overlaps the most recent one is ignored. Efficient 1 shows the empirically estimated probability that an HSP is
execution requires an array to record, for each diagonal, the first missed by this method, as a function of its normalized score. The
coordinate of the most recent hit found. Since database sequences two-hit method will detect an HSP if it contains two non-
are scanned sequentially, this coordinate always increases for overlapping length-W words of score at least T, with starting
successive hits. The idea of seeking multiple hits on the same positions that differ by no more than A residues. For W = 3, T = 11
diagonal was first used in the context of biological database and A = 40, Figure 1 shows the estimated probability that an HSP
searches by Wilbur and Lipman (19). is missed by this method, as a function of its normalized score. For
Because we require two hits rather than one to invoke an HSPs with score at least 33 bits, the two-hit heuristic is more
extension, the threshold parameter T must be lowered to retain sensitive.
comparable sensitivity. The effect is that many more single hits To analyze the relative speeds of the one-hit and two-hit
are found, but only a small fraction have an associated second hit methods, using the parameters studied above, we note that the
on the same diagonal that triggers an extension. The great two-hit method generates on average ∼3.2 times as many hits, but
majority of hits may be dismissed after the minor calculation of only ∼0.14 times as many hit extensions (Fig. 2). Because it takes
looking up, for the appropriate diagonal, the coordinate of the approximately one ninth as long to decide whether a hit need be
most recent hit, checking whether it is within distance A of the extended as actually to extend it, the hit-processing component of
current hit’s coordinate, and finally replacing the old with the new the two-hit method is approximately twice as rapid as the same
coordinate. Empirically, the computation saved by requiring component of the one-hit method.
fewer extensions more than offsets the extra computation
required to process the larger number of hits. TRIGGERING THE GENERATION OF GAPPED
To study the relative abilities of the one-hit and two-hit methods ALIGNMENTS
to detect HSPs of varying score, we model proteins using the
background amino acid frequencies of Robinson and Robinson Figure 1 shows that even when using the original one-hit method
(20), and use the BLOSUM-62 substitution matrix (18) for with threshold parameter T = 13, there is generally no greater than
sequence comparison. Given these Pi and sij, the statistical a 4% chance of missing an HSP with score >38 bits. While this
parameters for ungapped local alignments are calculated to be would appear sufficient for most purposes, the one-hit default T
λu = 0.3176 and Ku = 0.134. Using equation 3 above, we may parameter has typically been set as low as 11, yielding an
calculate the qij for which the scoring system is optimized, and execution time nearly three times that for T = 13. Why pay this
employ these target frequencies to generate model HSPs. Finally, price for what appears at best marginal gains in sensitivity? The
3392 Nucleic Acids Research, 1997, Vol. 25, No. 17

a b

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
c

Figure 3. A gapped extension generated by BLAST for the comparison of broad bean leghemoglobin I (87) and horse β-globin (88). (a) The region of the path graph
explored when seeded by the alignment of alanine residues at respective positions 60 and 62. This seed derives from the HSP generated by the leftward of the two
ungapped extensions illustrated in Figure 2. The Xg dropoff parameter is the nominal score 40, used in conjunction with BLOSUM-62 substitution scores and a cost
of 10 + k for gaps of length k. (b) The path corresponding to the optimal local alignment generated, superimposed on the hits described in Figure 2. The original BLAST
program, using the one-hit heuristic with T = 11, is able to locate three of the five HSPs included in this alignment, but only the first and last achieve a score sufficient
to be reported. (c) The optimal local alignment, with nominal score 75 and normalized score 32.4 bits. In the context of a search of SWISS-PROT (26), release 34
(21 219 450 residues), using the leghemoglobin sequence (143 residues) as query, the E-value is 0.54 if no edge-effect correction (22) is invoked. The original BLAST
program locates the first and last ungapped segments of this alignment. Using sum-statistics with no edge-effect correction, this combined result has an E-value of
31 (21,22). On the central lines of the alignment, identities are echoed and substitutions to which the BLOSUM-62 matrix (18) gives a positive score are indicated
by a ‘+’ symbol.

reason is that the original BLAST program treats gapped 0.95. The original algorithm, needing to find both HSPs, requires
alignments implicitly by locating, in many cases, several distinct 2P – P2 ≤ 0.05, or P less than ∼0.025. In contrast, the new
HSPs involving the same database sequence, and calculating a algorithm requires only that P2 ≤ 0.05, and thus can tolerate P as
statistical assessment of the combined result (21,22). This means high as 0.22. This permits the T parameter for the hit-stage of the
that two or more HSPs with scores well below 38 bits can, in algorithm to be raised substantially while retaining comparable
combination, rise to statistical significance. If any one of these sensitivity—from T = 11 to T = 13 for the one-hit heuristic. (The
HSPs is missed, so may be the combined result. two-hit heuristic described above lowers T back to 11.) As will be
The approach taken here allows BLAST to simultaneously discussed below, the resulting increase in speed more than
produce gapped alignments and run significantly faster than compensates for the extra time required for the rare gapped
previously. The central idea is to trigger a gapped extension for extension.
any HSP that exceeds a moderate score Sg, chosen so that no more In summary, the new gapped BLAST algorithm requires two
than about one extension is invoked per 50 database sequences.
non-overlapping hits of score at least T, within a distance A of one
(By equation 2, for a typical-length protein query, Sg should be set
another, to invoke an ungapped extension of the second hit. If the
at ∼22 bits.) A gapped extension takes much longer to execute
than an ungapped extension, but by performing very few of them HSP generated has normalized score at least Sg bits, then a gapped
the fraction of the total running time they consume can be kept extension is triggered. The resulting gapped alignment is reported
relatively low. only if it has an E-value low enough to be of interest. For example,
By seeking a single gapped alignment, rather than a collection in the pairwise comparison of Figure 2, the ungapped extension
of ungapped ones, only one of the constituent HSPs need be invoked by the hit pair on the left produces an HSP with score
located for the combined result to be generated successfully. This 23.6 bits (calculated using λu and Ku). This is sufficient to trigger
means that we may tolerate a much higher chance of missing any a gapped extension, which generates an alignment with score 32.4
single moderately scoring HSP. For example, consider a result bits (calculated using λg and Kg) and E-value of 0.5 (Fig. 3). The
involving two HSPs, each with the same probability P of being original BLAST program locates only the first and last ungapped
missed at the hit-stage of the BLAST algorithm, and suppose that segments of this alignment (Fig. 3c), and assigns them a
we desire to find the combined result with probability at least combined E-value >50 times greater.
 3393

Nucleic Acids
Nucleic Acids Research,
Research,1994,
1997,Vol.
Vol.22,
25,No.
No.117 3393

THE CONSTRUCTION AND STATISTICAL

EVALUATION OF GAPPED LOCAL ALIGNMENTS

The standard dynamic programming algorithms for pairwise

sequence alignment perform a fixed amount of computation per
cell of a path graph, whose dimensions are the lengths of the two
sequences being compared (23–25). In order to gain speed,
database search algorithms such as Fasta (2) and an earlier gapped
version of BLAST (3) sacrifice rigor by confining the dynamic
programming to a banded section of the full path graph (4),
chosen to include a region of already identified similarity. One
problem with this approach is that the optimal gapped alignment
may stray beyond the confines of the band explored. As the width
of the band is increased to reduce this possibility, the speed
advantage of the algorithm is vitiated.

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
We have accordingly taken a different heuristic approach to
constructing gapped local alignments, which is a simple general-
ization of BLAST’s method for constructing HSPs. The central Figure 4. The path graph region explored by BLAST during a gapped extension
idea is to consider only cells for which the optimal local alignment for the comparison of broad bean leghemoglobin I and the E1B protein small
score falls no more than Xg below the best alignment score yet T-antigen from human adenovirus type 4 (89) (SWISS-PROT accession no.
P10406). The Xg dropoff parameter is the nominal score 40, used in conjunction
found. Starting from a single aligned pair of residues, called the with BLOSUM-62 substitution scores and 10 + k gap costs. The 22.7 bit HSP
seed, the dynamic programming proceeds both forward and that triggers this extension, involving leghemoglobin residues 119–140 and
backward through the path graph (Zheng Zhang et al., manuscript adenovirus residues 101–122, is merely a random similarity, and not part of a
in preparation) (Figs 3a and 4). The advantage of this approach larger and higher-scoring alignment. The gapped extension is seeded by the
is that the region of the path graph explored adapts to the alignment of residues 124 and 106. The optimal alignment score through points
in the path graph drops steadily as one moves beyond the triggering HSP, and
alignment being constructed. The alignment can wander arbitrari- the reverse extension terminates before the beginning of either protein is
ly many diagonals away from the seed, but the number of cells reached. A total of 2766 path graph cells are explored, with the reverse
expanded on each row tends to remain limited, and may even extension accounting for 2047 of these cells.
shrink to zero before a boundary of the path graph is encountered
(Fig. 4). The Xg parameter serves a similar function to the
band-width parameter of the earlier heuristic, but the region of the path graph cells, so that a typical two-way gapped extension that
path graph it implicitly specifies be explored is in general more does not encounter the end of either sequence is expected to
productively chosen. involve ∼4000 cells. Because Sg is set so that a gapped extension
An important element for this heuristic is the intelligent choice is invoked less than once per 50 database sequences, fewer than
of a seed. Given an HSP whose score is sufficiently high that it 80 cells need be explored per database sequence.
triggers a gapped extension, how does one choose a residue pair The execution time required for a gapped extension is ∼500
to force into alignment? While more sophisticated approaches are times that for an ungapped extension. However, by triggering
possible, the simple procedure we have implemented is to locate, gapped extensions in the manner described, while simultaneously
along the HSP, the length-11 segment with highest alignment raising T for the single-hit version of BLAST from 11 to 13,
score, and use its central residue pair as the seed. If the HSP itself approximately one gapped extension is invoked for every 4000
is shorter than 11, a central residue pair is chosen. For example, ungapped extensions avoided. Because the number of ungapped
the first ungapped region in the alignment of Figure 3c constitutes extensions is reduced by about two thirds, the total time spent on
the HSP that triggered the alignment. The highest-scoring the extension stage of BLAST is cut by well over half. Of course,
length-11 segment of this HSP aligns leghemoglobin residues the two-hit strategy described above reduces the time needed for
55–65 with β-globin residues 57–67. Thus the alanine residues at the ungapped extensions still further. Once program overhead is
respective positions 60 and 62 are used as the seed for the gapped accounted for, the net speedup is a factor of about three.
extension illustrated in Figure 3a. As discussed in the perform- For any alignment actually reported, a gapped extension that
ance evaluation section below, this procedure is extremely good records ‘traceback’ information (25) needs to be executed. To
at selecting seeds that in fact participate in an optimal alignment. increase BLAST’s accuracy in producing optimal local align-
Most gapped extensions are triggered by chance similarities, ments, these gapped extensions use by default a substantially
and are therefore likely to be of limited extent, as illustrated in larger Xg parameter than employed during the program’s search
Figure 4. The reverse extension in this example explores ∼2000 stage.

Table 1. Relative times spent by the original and gapped BLAST programs on various algorithmic stages

Overhead: database Calculating whether hits Ungapped extensions Gapped extensions

scanning, output, etc. qualify for ungapped extension
Original BLAST 8 (8%) 92 (92%)
Gapped BLAST 8 (24%) 12 (37%) 5 (15%) 8 (24%)
3394 Nucleic Acids Research, 1997, Vol. 25, No. 17

The times required by various steps of the BLAST algorithm iteration takes little more than the same time to run. In related
vary substantially from one query and one database to another. work, Henikoff and Henikoff (39) have described how, short of
Table 1 shows typical relative times spent by the original and the modifying BLAST so that it may operate on a position-specific
gapped BLAST programs on various algorithmic stages. The score matrix, a single artificial sequence that approximates such
‘original BLAST’ program is represented, here and below, by a a matrix may be used as a query with the original BLAST
variant form of blastp version 1.4.9, modified so that it uses the programs.
same edge-effect correction (22) and background amino acid The construction of a position-specific score matrix is a
frequencies as the ‘gapped BLAST’. The times represent the multi-stage process, and at each stage a choice must be made
average for three different queries, with the time for the original among a number of alternative routes. We have been guided by
BLAST program normalized in each instance to 100 units. the goals of automatic operation, speed of execution, and general
More concretely, to search SWISS-PROT (26), release 34 simplicity. The issues discussed below are: (i) general architec-
(59 576 sequences; 21 219 450 residues), with the length-567 ture of the score matrix; (ii) construction of the multiple
influenza A virus hemagglutinin precursor (27) as query, the alignment from which the matrix is derived; (iii) weights for
original BLAST program requires 45.8 s, and the gapped BLAST sequences within the multiple alignment, and evaluation of the
program 15.8 s. This timing experiment, and others referred to effective number of independent observations it constitutes;
below, was run on one 200 MHz R10000 cpu processor of a (iv) estimation of target frequencies, and the construction of

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
lightly loaded SGI Power Challenge XL computer with 2.5 matrix scores; (v) applying BLAST to a position-specific matrix,
Gbytes of RAM. This machine runs the operating system IRIX, and the statistical evaluation of search results. We do not claim
version 6.2, which is an implementation of UNIX. We used the our current implementation is optimal, and it is likely that over
standard SGI C compiler, with the -O flag for optimization, to time some of its details will change.
compile all versions of the programs. The times reported are the
user times given by the time command, and are for the better of
two identical runs. Score matrix architecture
A closely related type of gapped extension routine to that used The alignment of a simple sequence with a pattern embodied by
here was developed by G. Myers during the evaluation of the a position-specific score matrix is almost completely analogous
original BLAST algorithm. It was not included in the publicly to the alignment of two simple sequences. The only real
distributed code primarily because the then current strategy of difference is that the score for aligning a letter with a pattern
extending every hit decreased the algorithm’s speed unduly for the position is given by the matrix itself, rather than with reference to
relatively small gain in sensitivity realized (1). a substitution matrix. For proteins, a query of length L and a
As discussed above, the statistical significance of gapped substitution matrix of dimension 20 × 20 are replaced by a
alignments may be evaluated using the two statistical parameters position-specific matrix of dimension L × 20. Position-specific
λg and Kg. The current version of the Fasta program (2) estimates gap costs may be defined as well (34,40). As with pairwise
these parameters on each run, by analyzing the distribution of sequence comparison, one may choose among finding the best
alignment scores produced by all the sequences in the database. global alignment of the matrix and the simple sequence (23),
BLAST gains speed by producing alignments for only the few finding the best alignment of the complete matrix with a segment
database sequences likely to be related to the query, and therefore of the sequence (41), and finding the best local alignment of the
does not have the option of estimating λg and Kg on the fly. matrix and sequence (24).
Instead, it uses estimates of these parameters produced before- Position-specific protein score matrices draw their power from
hand by random simulation (3). A drawback of this approach is two sources. The first is improved estimation of the probabilities
that the program may not accept an arbitrary scoring system, for with which amino acids occur at various pattern positions, leading
which no simulation has been performed, and still produce to a more sensitive scoring system. The second is relatively
accurate estimates of statistical significance. The original BLAST precise definition of the boundaries of important motifs. By
programs, in contrast, because they dealt only with ungapped demanding the complete alignment of one or more motifs, rather
local alignments, could derive λu and Ku from theory for any than seeking an arbitrary local alignment, the size of the search
scoring matrix (8,9). space may be greatly reduced, thereby lowering the level of
random noise. Unfortunately, there are many obstacles to
ITERATED APPLICATION OF BLAST TO automating well the delineation of a set of motifs from the output
POSITION-SPECIFIC SCORE MATRICES of a database search. The query sequence may contain a variety
of different domains, and share different subsets of them with
Database searches using position-specific score matrices, also different proteins in the database. Furthermore, defining the
called profiles or motifs, often are much better able to detect weak proper extent of even a single motif may be challenging (42).
relationships than are database searches that use a simple Accordingly, we have chosen to forgo the potential advantages
sequence as query (28–38). Employing these methods, however, of restricting the length of our derived matrices, and then
frequently has involved the use of several different programs and demanding that they be completely aligned with segments of
a fair degree of expertise. Accordingly, to render the power of database sequences (41). Instead, each matrix we construct has
motif searches more readily available, we have written a length precisely equal to that of the original query sequence.
procedure to construct a position-specific score matrix automati- When searching the database with such a matrix, we seek local
cally from the output of a BLAST run, and modified BLAST to alignments, in full analogy to those sought by BLAST when used
operate using such a matrix in the place of a simple query. The for straightforward sequence–sequence comparison. Finally, we
resulting PSI-BLAST program often is substantially more do not attempt to derive position-specific gap scores for use with
sensitive than the corresponding BLAST program, but for each our position-specific substitution scores. Instead, in each iteration
 3395

Nucleic Acids
Nucleic Acids Research,
Research,1994,
1997,Vol.
Vol.22,
25,No.
No.117 3395

a b

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
Figure 5. (a) The multiple alignment generated by PSI-BLAST when the human fragile histidine triad (HIT) protein (61) (SWISS-PROT accession no. P49789) is
compared to SWISS-PROT. All pairwise local alignments have E-value ≤0.01, and are identified in SWISS-PROT as belonging to the HIT family. Thick bars within
the six database sequences represent segments that align with various segments from the query. In constructing sequence weights for the indicated multiple alignment
column, corresponding to residue 108 of the query, only the shaded portions of the multiple alignment are used. (b) A local alignment of the human HIT protein and
H.influenzae galactose-1-phosphate uridylyltransferase (63) (SWISS-PROT accession no. P31764). In its first position-specific iteration, PSI-BLAST gives this
alignment a score of 45.4 bits, corresponding to an E-value of 4 × 10–5. ‘+’ symbols reflect positive BLOSUM-62 matrix scores, even though a position-specific matrix
is used to construct the alignment. (c) A local alignment of the human HIT protein and yeast 5′,5′′′-P1,P4-tetraphosphate phosphorylase I (64) (SWISS-PROT accession
no. P16550). In its second position-specific iteration, PSI-BLAST gives this alignment a score of 43.4 bits, corresponding to an E-value of 2 × 10–4.

of PSI-BLAST, we employ the same gap scores that are used in gap characters in every row and column (Fig. 5a), and is therefore
the first, simple BLAST run. Our reasons are that there is no good amenable to the various manipulations described below.
theory for deriving gap costs from a multiple alignment and that,
as will be discussed below, by eschewing position-specific gap Sequence weights
costs we can make a reasonable estimate of the statistical
significance of the resulting local alignments. When constructing a score matrix from a multiple alignment, it
is a mistake to give all sequences of the alignment equal weight.
Multiple alignment construction A large set of closely related sequences carries little more
information than a single member, but its size alone may allow it
To produce a multiple alignment from the BLAST output, we easily to ‘outvote’ a small number of more divergent sequences.
simply collect all database sequence segments that have been One way past this difficulty is to assign weights to the various
aligned to the query with E-value below a threshold, by default sequences, with those having many close relatives receiving
set to 0.01. The query is used as a master, or template, for smaller weight. The many sequence weighting methods that have
constructing a multiple alignment M. Any row (i.e., database been proposed (43–51) often produce roughly equivalent results.
sequence segment) identical to the query segment with which it Because of its speed and simplicity, we have implemented a
aligns is purged, and only one copy is retained of any rows that modified version of the sequence weighting method of Henikoff
are >98% identical to one another. Pairwise alignment columns and Henikoff (47). Gap characters are treated as a 21st distinct
that involve gap characters inserted into the query are simply character, and any columns consisting of identical residues are
ignored, so that M has exactly the same length as the query. ignored in calculating weights. In speaking of a column’s
Because we are dealing with local alignments, the columns of M observed residue frequencies fi, we shall henceforth mean its
may involve varying numbers of sequences, and many columns weighted rather its raw frequencies.
may include nothing but the query. We make no attempt to In constructing matrix scores, not only a column’s observed
improve M by comparing database sequences with one another, residue frequencies are important, but also the effective number
or by any other true multiple alignment procedure. of independent observations it constitutes: a column consisting of
As will be discussed, the matrix scores constructed for a given a single valine and a single isoleucine carries different informa-
alignment column should depend not only upon the residues tion than one consisting of five independently occurring instances
appearing there, but upon those in other columns as well. To make of each. Accordingly, we need to estimate the relative number NC
this dependency easy to formulate, however, we need to prune our of independent observations constituted by the alignment MC. A
raw multiple alignment M to a simpler ‘reduced’ one. This simple count of the number of sequences in MC is a poor measure,
pruning is done independently for each column, so the reduced for 10 identical sequences imply fewer independent observations
multiple alignment MC will in general vary from one column C than do 10 divergent ones. We thus propose as a simple first
to the next. To construct MC, we first specify the set R of estimate for NC the mean number of different residue types,
sequences it includes to be exactly those that contribute a residue including gap characters, observed in the various columns of MC.
to column C. We then define the columns of MC to be just those This estimate is clearly not ideal, as it saturates at 21 no matter
columns of M in which all the sequences of R are represented. By how many independent sequences are contained in MC. However,
construction, the reduced multiple alignment MC has residues or for the data we are likely to encounter, NC is typically much
3396 Nucleic Acids Research, 1997, Vol. 25, No. 17

smaller than 21, and therefore perhaps a good enough approxima- performed on a query consisting of a position-specific matrix
tion for our purposes. As will be seen, it is not the absolute value rather than a simple sequence. The same holds for the ungapped
of NC that is important, but rather its relative value from one and gapped extension steps of BLAST. One important issue is
column to another. NC is essentially the same measure of whether key parameters such as T and Xg, used at various heuristic
alignment variability as that proposed by Henikoff and Henikoff stages of the algorithm and tuned to simple sequence comparison,
(52) for use in a different manner. can be applied unchanged to position-specific matrices without
degrading unduly either the speed or sensitivity of database
Target frequency estimation searches. We approach this problem by ensuring that the scale λu
of the matrix scores produced internally by PSI-BLAST corre-
Given a multiple alignment, many methods for generating score sponds to that of the substitution matrix sij. In other words, we
matrices have been advanced (28–37,42,52–54). The prescrip- calculate the scores for a column of the matrix as [ln(Qi/Pi)]/λu.
tion with perhaps the best theoretical foundation is that the scores There is no analytic theory with which to estimate the statistical
for a specific pattern position be of the form log (Qi/Pi), where Qi significance of a gapped alignment of a position-specific score
is the estimated probability for residue i to be found in that column matrix and a simple sequence. However, one may hypothesize
(29,30,32,33,36,37,42,52–54). This leaves open the question that for a score matrix constructed to the same scale as sij, a given
how best to estimate the Qi. set of gap costs should produce the same gapped alignment scale

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
Given a multiple alignment involving a large number of parameter λg as for sij. This would be convenient, because then
independent sequences, the estimate of Qi for a specific column PSI-BLAST could estimate statistical significance without
should converge simply to the observed frequency of residue i in expending after each iteration the substantial time required to
that column. However, in addition to the sequence weighting estimate λg and Kg by random simulation. To test this hypothesis,
issues discussed above, factors that complicate estimating the Qi we performed a number of statistical tests on PSI-BLAST
include small sample size (30), and prior knowledge of relation- generated score matrices, scaled to have λu = 0.3176, the value
ships among the residues (16,37,53). Various studies suggest that applicable to previously published BLOSUM-62 simulations (3).
the best currently available method for estimating the Qi is that of First, we searched SWISS-PROT using as query the length-567
Dirichlet mixtures (52–56). However, because it often performs influenza A virus hemagglutinin precursor (27), and captured the
nearly as well (52), and due to its relative simplicity, we have score matrix constructed by PSI-BLAST from the 128 local
implemented the data-dependent pseudocount method intro- alignments with E-value ≤ 0.01. We then compared this matrix to
duced by Tatusov et al. (37). This method uses the prior 10 000 random sequences of length 567, generated using the
knowledge of amino acid relationships embodied in the substitu- background amino acid frequencies of Robinson and Robinson
tion matrix sij to generate residue pseudocount frequencies gi, (20). A gap of length k was charged a cost of 10 + k. Counts of
which are averaged with the observed frequencies fi to estimate the optimal local alignment scores, calculated using an appropri-
the Qi. ately modified version of the Smith–Waterman algorithm (24),
Specifically, for a given column C, we construct pseudocount are plotted in Figure 6. Also shown is the best fitting extreme
frequencies gi using the formula: value distribution (3,15) which, using the edge-effect correction
described by Altschul and Gish (3), has statistical parameters λg
gi   Pf q
j

j
ij 4 = 0.251 and Kg = 0.031. It is apparent that the distribution fits the
random trial reasonably well; a χ2 goodness-of-fit test with 34
j
degrees of freedom has value 41.8, which is lower than one would
where the qij are the target frequencies implicit in the substitution expect 20% of the time even were the theory precisely valid. This
matrix, and given by equation 3. Intuitively, those residues supports the idea that the statistical theory described above
favored by the substitution matrix to align with the residues applies to local alignments of position-specific score matrices and
actually observed receive high pseudocount frequencies. We then simple sequences. Furthermore, the estimate λg ≈ 0.251 ± 0.003
estimate Qi by: agrees to within experimental error with the value 0.255
previously published for these gap costs (3). Similar agreement
af i
bg i
Qi  5 was obtained with a number of other protein sequences as initial
a
b query (results not shown), and in all cases the much less important
where α and β are the relative weights given to observed and Kg parameter could be estimated accurately as well. In general,
pseudocount residue frequencies. So that the scores we construct values of λg for the comparison of position-specific score
will reduce to sij in columns where nothing has been aligned to the matrices with simple sequences appear to differ by <2% from the
query sequence, we let α = NC – 1. β remains an arbitrary values for simple pairwise sequence comparison. Using these
pseudocount parameter; the larger its value, the greater the precomputed values for λg should thus entail an error of less than
emphasis given to prior knowledge of residue relationships vis a one bit for PSI-BLAST scores <50 bits, corresponding to a factor
vis observed residue frequencies. We have found empirically that, of less than two in the estimation of statistical significance.
in conjunction with our method for calculating α, a reasonably
good setting for β is 10. PERFORMANCE EVALUATION

BLAST applied to position-specific score matrices To test more directly the statistics used by PSI-BLAST, we
compared query sequences from 11 large and well characterized
The initial step of the BLAST algorithm is the construction of a protein families to the SWISS-PROT database, and then ran the
list of words that align to query words with score at least T. Only position-specific score matrices generated against a shuffled
minor modifications to the code are necessary for this step to be version of the same database. For each query, we recorded the
 3397

Nucleic Acids
Nucleic Acids Research,
Research,1994,
1997,Vol.
Vol.22,
25,No.
No.117 3397

identical shuffled-database test on the gapped and original

versions of BLAST. To reduce the probability that high-scoring
alignments were missed due to the heuristic nature of the
algorithms, we performed these tests with T = 9 rather than the
default value of 11. The results are given in Table 2. For the 11
queries, the median of the low PSI-BLAST E-values was 0.87,
which corresponds to a median P-value of 0.58 (8,9). The mean
numbers of shuffled database sequences with E-values <1 and 10
were 1.0 and 8.7, respectively, within 20% of the expected values
of 1.0 and 10.0. The equivalent tests for the ungapped and gapped
versions of BLAST also yielded results that diverged from theory
by <50%.
The ability to estimate with reasonable accuracy the signifi-
cance of gapped local matrix-sequence alignments permits us to
automate the construction of position-specific score matrices

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
during multiple iterations of the PSI-BLAST program. After each
iteration, we generate a new multiple alignment simply by
collecting those alignments with E-value lower than a defined
Figure 6. The distribution of optimal local alignment scores from the threshold. An interactive version of PSI-BLAST allows the user
comparison of a position-specific score matrix with 10 000 random protein to override either the inclusion or exclusion of specific local
sequences. The score matrix was constructed by PSI-BLAST from the 128 local
alignments with E-value ≤0.01 found in a search of SWISS-PROT using as alignments. Once a given database sequence has been used in the
query the length-567 influenza A virus hemagglutinin precursor (27) (SWISS- generation of a position-specific score matrix, low E-values for
PROT accession no. P03435). The random sequences, each of length 567, were this sequence are virtually guaranteed in future iterations, for the
generated using the amino acid frequencies of Robinson and Robinson (20). sequence is to a certain extent being compared with itself. The
Optimal local alignment scores were calculated using the position-specific
matrix in conjunction with 10 + k gap costs. The extreme value distribution that biological relevance of PSI-BLAST output thus depends criti-
best fits the data (3,15) is plotted. A χ2 goodness-of-fit test with 34 degrees of cally on avoiding the inappropriate inclusion of sequences in the
freedom has value 41.8, corresponding to a P-value of 0.20. multiple alignment constructed. Specifically, the utility of the
score matrix produced is immediately vitiated by the inclusion of
lowest E-value found, as well as the number of shuffled sequences any alignment involving a region of highly biased amino acid
yielding E-values ≤1 and 10. For comparison, we performed the composition (57,58).

Table 2. The comparison of various query sequences with a shuffled version of SWISS-PROT

Protein family SWISS-PROT Original BLAST Gapped BLAST PSI-BLAST

accession no. Low No. of seqs Low No. of seqs Low No. of seqs
of query E-value with E-value E-value with E-value E-value with E-value
≤1 ≤10 ≤1 ≤10 ≤1 ≤10
Serine protease P00762 0.86 1 7 3.0 0 4 0.94 1 8
Serine protease inhibitor P01008 3.9 0 4 0.078 1 9 1.5 0 9
Ras P01111 3.4 0 8 3.4 0 7 1.1 0 9
Globin P02232 2.4 0 7 2.8 0 5 8.2 0 2
Hemagglutinin P03435 0.11 2 11 0.46 3 16 0.87 1 8
Interferon α P05013 2.4 0 6 0.27 2 4 0.11 2 11
Alcohol dehydrogenase P07327 1.5 0 2 0.80 1 5 1.5 0 9
Histocompatibility antigen P10318 0.91 1 7 0.13 1 7 0.0031 2 6
Cytochrome P450 P10635 0.84 2 5 8.5 0 3 0.46 1 15
Glutathione transferase P14942 1.0 1 10 3.3 0 3 0.30 2 9
H+-transporting ATP synthase P20705 0.012 1 8 0.26 2 14 0.79 2 10

Average (median or mean) 1.0 0.7 6.8 0.80 0.9 7.0 0.87 1.0 8.7

The original and gapped BLAST comparisons use BLOSUM-62 substitution scores (18). All three programs use threshold T parameter set to 9, but the gapped
BLAST and PSI-BLAST programs use the two-hit method to trigger ungapped extensions. The original BLAST program has the X dropoff parameter set to nominal
score 23. The gapped BLAST and PSI-BLAST comparisons charge gaps of length k a cost of 10 + k. They have Xu set to 16, and Xg set to 40 for the database search
stage and to 67 for the output stage of the algorithms. Gapped alignments are triggered by a score corresponding to ∼22 bits. For PSI-BLAST, the query is first com-
pared to the SWISS-PROT database, and the position-specific score matrix generated is then compared to a shuffled version of SWISS-PROT. The median is used
for the average of the low E-values, and the mean otherwise.
3398 Nucleic Acids Research, 1997, Vol. 25, No. 17

Table 3. The number of SWISS-PROT sequences yielding alignments with E-value ≤0.01, and relative running times, for Smith–Waterman and various versions
of BLAST

Protein family Query Smith–Waterman Original BLAST Gapped BLAST PSI-BLAST

Serine protease P00762 275 273 275 286
Serine protease inhibitor P01008 108 105 108 111
Ras P01111 255 249 252 375
Globin P02232 28 26 28 623
Hemagglutinin P03435 128 114 128 130
Interferon α P05013 53 53 53 53
Alcohol dehydrogenase P07327 138 128 137 160
Histocompatibility antigen P10318 262 241 261 338
Cytochrome P450 P10635 211 197 211 224
Glutathione transferase P14942 83 79 81 142
H+-transporting ATP synthase P20705 198 191 197 207

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
Normalized running time 36 1.0 0.34 0.87

To score and evaluate the significance of the alignments found, the original BLAST program uses BLOSUM-62 substitution scores (18) and sum-statistics (21,22).
The Smith–Waterman and gapped BLAST programs use BLOSUM-62 substitution scores, 10 + k gap costs, and the statistics of equations 1 and 2, in conjunction
with the experimentally determined parameters λg = 0.255 and Kg = 0.035 (3). PSI-BLAST uses the same gap costs and λg, but applied to the position-specific score
matrix constructed from the output of the gapped BLAST run. Only one PSI-BLAST iteration is executed. All three BLAST programs use the same parameter settings
as in Table 2, except that T is set to 11. Normalized running times are the mean ratio of program running time to that for the original BLAST. The time for PSI-BLAST
includes the time for the initial BLAST search.

To compare the performance of the new gapped version of A search that includes a single PSI-BLAST iteration still runs
BLAST and its PSI-BLAST extension to that of the Smith– faster than the original BLAST, and 40 times faster than
Waterman algorithm (24) and the original ungapped BLAST Smith–Waterman, but can in many cases be much more sensitive.
algorithm, we employed the same 11 query sequences that were It finds every true positive returned by Smith–Waterman, but
used above to investigate the accuracy of PSI-BLAST statistics. frequently many others as well. Here only a single PSI-BLAST
Because, as shown, these statistics are quite accurate, we may use iteration has been considered but, as will be seen below, multiple
the number of statistically significant sequences found in a iterations can yield even better results. Furthermore, we have
database search as a reasonable measure of algorithm sensitivity. found PSI-BLAST to perform better on searches of the non-
We employed the ssearch program, version 2.0u54, from the redundant protein sequence database maintained by the NCBI
Fasta package (2) as our implementation of the Smith–Waterman (59) than on searches of SWISS-PROT, because of the greater
algorithm. Using each of the 11 queries, we searched SWISS- number of significant similarities that are found by the initial
PROT with each of the four programs. We show in Table 3 the BLAST run.
numbers of sequences found with E value ≤0.01, as well as the For the particular examples in Table 3, the PSI-BLAST
average ratio of running time to that for the original BLAST iteration takes noticeably longer than the gapped BLAST
program. Based upon SWISS-PROT annotation, all sequences iteration, due primarily to the time needed to construct the
recorded in Table 3 appear to be true family members, with the position-specific score matrix from the large number of signifi-
exception of one of the lowest-scoring alignments found by cant local alignments found by BLAST. For queries that return a
Smith–Waterman when applied to the histocompatibility antigen small number of significant alignments, each PSI-BLAST
query, and the lowest-scoring alignment found by the original iteration requires more nearly the same time as BLAST.
BLAST applied to the hemagglutinin query. While some
alignments involve hypothetical proteins, the pattern of con- PSI-BLAST EXAMPLES
served residues in all such cases suggests a true positive.
As can be seen, the gapped BLAST program runs on average In many instances, PSI-BLAST is able automatically to uncover
three times faster then the original, and in all but one case biologically interesting similarities that elude simple database
examined finds a greater number of statistically significant searches. Multiple iterations of PSI-BLAST are sometimes required
alignments. It runs >100 times faster than Smith–Waterman, but to recognize the more distantly related protein family members. We
for the combined 11 queries misses only eight of the 1739 here consider two representative cases in greater detail.
significant similarities found by the rigorous algorithm. Of these
eight, only one has an E-value <0.001, and another appears to be HIT proteins
a random as opposed to a biologically meaningful similarity. The
scores produced by gapped BLAST for the 1731 similarities it Holm and Sander (60) describe how a comparison of three-dimen-
finds differ from those produced by the Smith–Waterman sional structures is able to identify significant similarity between
algorithm in only two instances. The discrepancy arises in both histidine triad (HIT) proteins and galactose-1-phosphate uridylyl-
cases from an Xg parameter that is too low rather than from an transferase (GalT) proteins. Indeed, using the human HIT protein
incorrect choice of seed. Thus despite its simplicity, the (61) as query, a BLAST search of SWISS-PROT reveals hits with
seed-selection heuristic is extremely accurate. E-value <0.01 only to other HIT proteins (Fig. 5a). An alignment to
 3399

Nucleic Acids
Nucleic Acids Research,
Research,1994,
1997,Vol.
Vol.22,
25,No.
No.117 3399

the rat GalT protein (62) has the only marginally significant E-value
of 0.012. A PSI-BLAST search, using the score matrix generated
from the six alignments illustrated in Figure 5a, can immediately
cement confidence in the biological relevance of this similarity. The
E-value of the similarity with rat GalT drops to 2 × 10–4, and an
alignment to Haemophilus influenzae GalT (63) (Fig. 5b) receives
the even more significant E-value of 4 × 10–5. These similarities, of
course, are uncovered using no structural information. In addition,
on the next iteration, PSI-BLAST finds a strongly significant
alignment (Fig. 5c; E-value 2 × 10–4) to yeast 5′,5′′′-P1,P4-tetra-
phosphate phosphorylase I (64), for which no structure is available.

BRCT proteins
Proteins containing one or multiple copies of the BRCT domain
Figure 7. The location of BRCT domains within human BRCA1 (68),

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
form a superfamily many of whose members are involved in DNA Arabidopsis T10M13.12 (72), human KIAA0259 (73), worm T13F2.3 (74),
damage-responsive cell cycle checkpoints (65–67). While detailed fission yeast SPAC6G9.12 (75), worm C36A4.8 (74), Synechocystis sp.
analysis is needed to delineate completely this diverse set of D90904 (77) and human Pescadillo (79). BRCA1 and C36A4.8 each have, in
proteins, PSI-BLAST is able to automatically identify most of the addition, an N-terminal RING finger domain. The near identity to other worm
sequences of a short region directly preceding the BRCT domain of C36A4.8
superfamily. We used the C-terminal 215 residues of human suggests the possibility that this protein has been misassembled.
BRCA1 (68), which includes two BRCT domains (65), as the
initial query for a search of NCBI’s non-redundant protein
sequence database. Using the default cutoff E-value of 0.01, the (vii) Pescadillo is a human protein whose zebrafish ortholog is
initial BLAST search recognized as significant only alignments to essential for embryonic development (79), and whose yeast
other BRCA1 sequences, and the previously described BRCT ortholog YGR103w (80) has been previously recognized as a
protein BARD (69) (Table 4). Subsequent PSI-BLAST iterations, BRCT protein (66,67). It failed to pass the cutoff E-value of 0.01,
however, retrieved all the proteins recorded in Table 4; additional but appeared with near-significant E-values in PSI-BLAST output
close homologs are omitted from the table. Almost all the BRCT from the 5th iteration onward. The approximate positions of the
proteins described by Bork et al. (66) were recognized. Not found BRCT domains within BRCA1 and the seven newly identified
were the retinoblastoma family, whose putative BRCT domain is BRCT proteins are illustrated in Figure 7.
particularly divergent, worm R13A5.13, which was not in the
database searched, and human DNA-ligase III. PSI-BLAST did DISCUSSION AND CONCLUSION
report yeast RAD9 and YGR103w, the Kluyveromyces lactis
RAP1 homolog, worm ZK675.2, and various terminal deoxynu- In addition to the major algorithmic changes described above, we
cleotidyltransferases and poly(ADP-ribose) polymerases, but all have modified an aspect of the original BLAST program’s output
with E-values >0.01 (Table 4). Detailed examination of the routine that on occasion caused important similarities to be
alignments produced suggests that the only likely false positives overlooked. When a very large number of statistically significant
involved a trypanosome EST (70) and the Methanococcus alignments was found, BLAST would typically report only the top
jannaschii mutT protein (71), the latter despite its involvement in scoring 500. These alignments, however, might all involve one
DNA repair (Table 4). domain of the query that occurred frequently within the database.
Seven recent additions to the protein databases are reported here Interesting but weaker relationships to other regions of the query
as members of the BRCT superfamily (Table 4). (i) Arabidopsis might simply be forced off the bottom of the list. Accordingly,
T10M13.12 (72), is the first plant protein observed to contain following the general idea of Sonnhammer and Durbin (81), we
BRCT domains. (ii) KIAA0259 (73) is a large human protein of have limited the number of alignments reported that involve each
unknown function with eight BRCT domains, the greatest number region of the query, but set no overall upper limit.
so far observed within a single protein. (iii) T13F2.3 (74) is a worm The BLAST programs are unlikely to remain static, and there
protein with a 500-residue low-complexity (57) N-terminus. are many possible avenues for future improvement. We discuss
(iv) SPAC6G9.12 (75) is a fission yeast protein strongly similar to three of them briefly here.
the previously recognized (66) yeast BRCT protein L8543.18 (76).
(v) C36A4.8 (74) is a worm protein whose C-terminus contains a Gap costs
single BRCT domain, and whose N-terminus, containing a RING
finger domain, is strongly similar to that of BRCA1. The similar Gapped alignments may be constructed using a variety of different
organization to BRCA1 makes this protein of particular interest. types of gap cost. Because a single mutational event may insert or
(vi) Synechocystis sp. D90904 (77) is the first bacterial BRCT delete a large number of residues, it has been argued that long gaps
protein that is not a bacterial ligase. While it failed to pass the cutoff should not cost much more than short ones, and affine gap costs,
E-value of 0.01, its C-terminal BRCT domain is very similar to that which assess a score –(a + bk) for a gap of length k (82–85), have
of several bacterial ligases, which presumably led to its incorrect become the most widely used. A generalization of these costs has
classification as such in the databases. Most of the protein been proposed, that allows a gap to involve residues in both
N-terminal to its BRCT domain consists of a coiled-coil domain. sequences rather than just one (86). Specifically, a gap in which k
The actual Synechocystis sp. DNA ligase (77,78) is found by residues are inserted or deleted and j pairs of residues are left
PSI-BLAST on the 13th iteration, with an E-value of 0.002. unaligned receives the score –(a + bk + cj). The algorithm necessary
3400 Nucleic Acids Research, 1997, Vol. 25, No. 17

for using such costs is only a minor variant on that for traditional it is justified, they may lead to the construction of more sensitive
affine gap costs. In many cases, the new gap costs generate local position-specific score matrices. Whether it is desirable to use
alignments that are both more accurate and more statistically generalized affine gap costs as the default for general purpose
significant (86). These costs are potentially of particular value for database searches awaits detailed empirical study.
use with PSI-BLAST, because by imposing alignment only where

Table 4. PSI-BLAST protein database search results using the C-terminus of BRCA1 as query

Protein Species GenBank ID number PSI-BLAST iteration E-value

BARD Homo sapiens 1710175 0 2e-06
T10M13.12a Arabidopsis thaliana 2104545 1 4e-06
F26D2.bb Caenorhabditis elegans 1914176 1 4e-04
KIAA0259a H.sapiens 1665785 1 0.001
F37D6.1 C.elegans 1418521 2 4e-06
C19G10.07 Schizosaccharomyces pombe 1723501 2 6e-05

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
KIAA0170 H.sapiens 1136400 2 0.002
53BP1 H.sapiens 488592 2 0.008
T13F2.3a C.elegans 1667334 3 2e-07
K04C2.4 C.elegans 470351 3 3e-07
T19E10.1 C.elegans 1067065 4 7e-04
Rad4/Cut5 S.pombe 730470 4 0.002
REV1 Saccharomyces cerevisiae 132409 4 0.003
ECT2 Mus musculus 423597 5 1e-04
XRCC1 M.musculus 627867 5 6e-04
Crb2 S.pombe 1449177 5 0.002
RAP1 S.cerevisiae 173558 5 0.006
TcEST030c Trypanosoma cruzi 1536857 6 0.001
DPB11 S.cerevisiae 1352999 6 0.001
L8543.18 S.cerevisiae 1078075 6 0.010
SPAC6G9.12a S.pombe 1644324 7 4e-04
YM8021.03 S.cerevisiae 1078533 7 0.005
YHR154w S.cerevisiae 731729 7 0.008
C36A4.8a C.elegans 1657667 7 0.010
UNE452 S.cerevisiae 1151000 8 8e-04
DNA ligase IV H.sapiens 1706482 8 0.008
CDC9 Candida albicans 1706483 9 0.006
DNA ligase Thermus scotoductus 1352293 10 0.010
GNF1 Drosophila melanogaster 544404 11 0.004
mutTc M.jannaschii 2129134 15 0.008

RAD9 S.cerevisiae 131817 7 0.74

RAP1 homolog K.lactis 422087 9 0.21
ZK675.2 C.elegans 599712 13 3.5
D90904a Synechocystis sp. 1652299 15 0.17
TDT Mus domestica 2149634 15 0.46
YGR103w S.cerevisiae 1723693 16 0.017
Pescadilloa H.sapiens 2194203 16 0.017
PPOL Sarcophaga peregrina 1709741 16 0.060

Iteration zero refers to the initial BLAST run, using the 215 C-terminal residues of BRCA1 (68) (SWISS-PROT accession no.
P38398) as query. Subsequent PSI-BLAST iterations use derived position-specific score matrices in place of the query. The score
matrix for iteration i + 1 is constructed from alignments achieving an E-value ≤0.01 for iteration i. For each protein, the E-value is
that returned during the PSI-BLAST iteration indicated, and precedes the protein’s use for score matrix construction. Only one repre-
sentative is listed for families of closely related proteins. On its 16th iteration PSI-BLAST uncovered no new proteins with E-value
≤0.01, and therefore ceased iteration. At the end of the table are shown BRCT proteins returned by PSI-BLAST with E-value >0.01
but ≤10, listed for the iteration in which they achieved their lowest E-value.
aRecent additions to the database, first identified as BRCT proteins here.
bThe C.elegans F26D2.b protein (74) while a recent addition to the databases, is a close homolog of the previously recognized (66,67)
family of C.elegans BRCT proteins containing, for example, F37A4.4 (90).
cThe trypanosome EST (70) and the M.jannaschii mutT protein (71) are the only likely false positives.
 3401

Nucleic Acids
Nucleic Acids Research,
Research,1994,
1997,Vol.
Vol.22,
25,No.
No.117 3401

Position-specific score matrices as input to PSI-BLAST retains the ability to report accurate statistics, per iteration runs in
times not much greater than gapped BLAST, and can be used both
PSI-BLAST performs three distinct operations: it constructs a iteratively and fully automatically. These developments should
multiple alignment from BLAST output data; it processes this enhance significantly the utility of database search methods to the
alignment into a position-specific score matrix; and it uses this molecular biologist.
matrix to search the database. A researcher may wish, however, to
bypass the first two of these operations, and provide a score matrix
as query directly to PSI-BLAST. The central difficulty is retaining Note
the ability to calculate reliable statistics; as described above, Source code for the new BLAST programs is available by
PSI-BLAST imposes strict scaling rules on the matrices it generates, anonymous ftp from the machine ncbi.nlm.nih.gov, within the
permitting the use of precomputed λg to assess significance. Three directory ‘blast’, and the programs may be run from NCBI’s web
possible routes are open. (i) One may permit the specification of 20 site at http://www.ncbi.nlm.nih.gov/
target frequencies Qi for each position rather than 20 scores. These
can then be converted internally to log-odds scores with the ACKNOWLEDGEMENTS
appropriate scale for precomputed statistical parameters to apply.
(ii) One may estimate by random simulation the statistical para- W.M. and Z.Z. are supported by grant LM05110 from the

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
meters for an input score matrix (3). An advantage is the National Library of Medicine. We thank Dr Warren Gish for
applicability to a greater range of scoring systems, including the helpful conversations, Dr Eugene Koonin for assistance with the
possible use of position-specific gap costs. A disadvantage is that examples, and Dr Gregory Schuler for producing several of the
obtaining reasonably accurate estimates of the relevant statistical figures.
parameters may increase the program’s execution time unduly.
(iii) One may abandon the statistical assessment of the alignments
produced. This gives the greatest scope to input scoring systems, but REFERENCES
precludes any reasonable scheme for automatic iteration of PSI- 1 Altschul,S.F., Gish,W., Miller,W., Myers,E.W. and Lipman,D.J. (1990)
BLAST. Much experimentation will be required to determine which J. Mol. Biol., 215, 403–410.
of these approaches is the most fruitful. 2 Pearson,W.R. and Lipman,D.J. (1988) Proc. Natl. Acad. Sci. USA, 85,
2444–2448.
3 Altschul,S.F. and Gish,W. (1996) Methods Enzymol., 266, 460–480.
Realignment 4 Chao,K.-M., Pearson,W.R. and Miller,W. (1992) Comput. Appl. Biosci., 8,
481–487.
After the initial BLAST run, or a later PSI-BLAST iteration, the 5 Sellers,P.H. (1984) Bull. Math. Biol., 46, 501–514.
multiple alignment used for subsequent iterations can be con- 6 Altschul,S.F. and Erickson,B.W. (1986) Bull. Math. Biol., 48, 633–660.
structed in a more sophisticated manner than described above. 7 Waterman,M.S. and Eggert,M. (1987) J. Mol. Biol., 197, 723–728.
8 Karlin,S. and Altschul,S.F. (1990) Proc. Natl. Acad. Sci. USA, 87,
Rather than coalescing all the pairwise alignments that pass the 2264–2268.
threshold immediately into a multiple alignment, the most 9 Dembo,A., Karlin,S. and Zeitouni,O. (1994) Ann. Prob., 22, 2022–2039.
significant among them can be used to build an initial multiple 10 Altschul,S.F. (1991) J. Mol. Biol., 219, 555–565.
alignment and associated position-specific score matrix, which 11 Altschul,S.F. (1993) J. Mol. Evol., 36, 290–300.
can then be used to rescore and realign database sequences that 12 Smith,T.F., Waterman,M.S. and Burks,C. (1985) Nucleic Acids Res., 13,
645–656.
received lower scores. This step can be iterated several times 13 Collins,J.F., Coulson,A.F.W. and Lyall,A. (1988) Comput. Appl. Biosci., 4,
before another full-scale database search is executed. There are 67–71.
several potential advantages to this procedure. (i) Weaker 14 Mott,R. (1992) Bull. Math. Biol., 54, 59–75.
pairwise alignments, that may be somewhat inaccurate, can be 15 Waterman,M.S. and Vingron,M. (1994) Stat. Sci., 9, 367–381.
16 Dayhoff,M.O., Schwartz,R.M. and Orcutt,B.C. (1978) In Dayhoff,M.O.
improved and perhaps extended before they are incorporated into (ed.), Atlas of Protein Sequence and Structure. National Biomedical
the evolving multiple alignment. (ii) Unrelated sequences that Research Foundation, Washington, DC, Vol. 5, Suppl. 3, pp. 345–352.
received chance high scores can have their scores downgraded by 17 Schwartz, R.M. and Dayhoff,M.O. (1978) In Dayhoff,M.O. (ed.), Atlas of
an improved matrix, and perhaps be rejected before they are Protein Sequence and Structure. National Biomedical Research
included in the alignment. (iii) Related sequences that received Foundation, Washington, DC, Vol. 5, Suppl. 3, pp. 353–358.
18 Henikoff,S. and Henikoff,J.G. (1992) Proc. Natl. Acad. Sci. USA, 89,
relatively high alignment scores, but missed the threshold for 10915–10919.
inclusion, can have their scores increased, and perhaps be 19 Wilbur,W.J. and Lipman,D.J. (1983) Proc. Natl. Acad. Sci. USA, 80,
included in the multiple alignment. In short, the realignment 726–730.
procedure can prevent inaccurate pairwise alignments from 20 Robinson,A.B. and Robinson,L.R. (1991) Proc. Natl. Acad. Sci. USA, 88,
8880–8884.
corrupting the evolving multiple alignment, and can accelerate 21 Karlin,S. and Altschul,S.F. (1993) Proc. Natl. Acad. Sci. USA, 90,
the recognition of related sequences, all for very little computa- 5873–5877.
tional cost. Preliminary studies suggest this line of development 22 Altschul,S.F. (1997) In Suhai,S. (ed.), Theoretical and Computational
to be a fruitful one. Methods in Genome Research. Plenum Press, New York, pp. 1–14.
In conclusion, the new gapped version of BLAST is both 23 Needleman,S.B. and Wunsch,C.D. (1970) J. Mol. Biol., 48, 443–453.
24 Smith,T.F. and Waterman,M.S. (1981) J. Mol. Biol., 147, 195–197.
considerably faster than the original one, and able to produce 25 Sankoff,D. and Kruskal,J.B. (1983) Time Warps, String Edits and
gapped alignments. While the relevant statistical parameters can Macromolecules: The Theory and Practice of Sequence Comparison.
no longer be calculated from theory, random simulation allows Addison-Wesley, Reading, MA.
them to be estimated beforehand for commonly used amino acid 26 Bairoch,A. and Apweiler,R. (1997) Nucleic Acids Res., 25, 31–36.
27 Jou,W.M., Verhoeyen,M., Devos,R., Saman,E., Fang,R., Huylebroeck,D.,
substitution matrices and gap costs. For many queries, the Fiers,W., Threlfall,G., Barber,C., Carey,N. and Emtage,S. (1980) Cell, 19,
PSI-BLAST extension can greatly increase sensitivity to weak 683–696.
but biologically relevant sequence relationships. PSI-BLAST 28 McLachlan,A.D. (1983) J. Mol. Biol., 169, 15–30.
3402 Nucleic Acids Research, 1997, Vol. 25, No. 17

29 Staden,R. (1984) Nucleic Acids Res., 12, 505–519. 61 Ohta,M., Inoue,H., Cotticelli,M.G., Kastury,K., Baffa,R., Palazzo,J.,
30 Schneider,T.S., Stormo,G.D., Gold,L. and Ehrenfeucht,A. (1986) J. Mol. Siprashvili,Z., Mori,M., McCue,P., Druck,T., Croche,C.M. and Huebner,K.
Biol., 188, 415–431. (1996) Cell, 84, 587–597.
31 Taylor,W.R. (1986) J. Mol. Biol., 188, 233–258. 62 Heidenreich,R.A., Mallee,J. and Segal,S. (1993) DNA Seq., 3, 311–318.
32 Berg,O.G. and von Hippel,P.H. (1987) J. Mol. Biol., 193, 723–750. 63 Maskell,D.J., Szabo,M.J., Deadman,M.E. and Moxon,E.R. (1992) Mol.
33 Dodd,I.B. and Egan,J.B. (1987) J. Mol. Biol., 194, 557–564. Microbiol., 6, 3051–3063.
34 Gribskov,M., McLachlan,A.D. and Eisenberg,D. (1987) Proc. Natl. Acad. 64 Plateau,P., Fromant,M., Schmitter,J.M., Buhler,J.M. and Blanquet,S. (1989)
Sci. USA, 84, 4355–4358. J. Bacteriol., 171, 6437–6445.
35 Patthy,L. (1987) J. Mol. Biol., 198, 567–577. 65 Koonin,E.V., Altschul,S.F. and Bork,P. (1996) Nature Genet., 13, 266–268.
36 Stormo,G.D. and Hartzell,G.W. III (1989) Proc. Natl. Acad. Sci. USA, 86, 66 Bork,P., Hofmann,K., Bucher,P., Neuwald,A.F., Altschul,S.F. and
1183–1187. Koonin,E.V. (1997) FASEB J., 11, 68–76.
37 Tatusov,R.L., Altschul,S.F. and Koonin,E.V. (1994) Proc. Natl. Acad. Sci. 67 Callebaut,I. and Mornon,J.-P. (1997) FEBS Lett., 400, 25–30.
USA, 91, 12091–12095. 68 Miki,Y., Swensen,J., Shattuck-Eidens,D., Futreal,P.A., Harshman,K.,
Tavtigian,S., Liu,Q., Cochran,C., Bennett,L.M., Ding,W., et al. (1994)
38 Yi,T.-M. and Lander,E.S. (1994) Protein Sci., 3, 1315–1328.
Science, 266, 66–71.
39 Henikoff,S. and Henikoff,J.G. (1997) Protein Sci., 6, 698–705.
69 Wu,L.C., Wang,Z.W., Tsan,J.T., Spillman,M.A., Phung,A., Xu,X.L.,
40 Bucher,P., Karplus,K., Moeri,N. and Hofmann,K. (1996) Comput. Chem.,
Yang,M.C., Hwang,L.Y., Bowcock,A.M. and Baer,R. (1996) Nature
20, 3–23. Genet., 14, 430–440.
41 Sellers, P.H. (1980) J. Algorithms, 1, 359–373. 70 Tanaka,T. and Tanaka, M. (1996) DDBJ accession no. D87228.
42 Lawrence,C.E., Altschul,S.F., Boguski,M.S., Liu,J.S., Neuwald,A.F. and

Downloaded from http://nar.oxfordjournals.org/ at UNAM Direccion General de Bibliotecas on August 11, 2014
71 Bult,C.J., White,O., Olsen,G.J., Zhou,L., Fleischmann,R.D., Sutton,G.G.,
Wootton,J.C. (1993) Science, 262, 208–214. Blake,J.A., FitzGerald,L.M., Clayton,R.A., Gocayne,J.D., et al. (1996)
43 Altschul,S.F., Carroll,R.J. and Lipman,D.J. (1989) J. Mol. Biol., 207, Science, 273, 1058–1073.
647–653. 72 Johnson,A.F., de la Bastide,M., Lodhi,M., Hoffman,J., Hasegawa,A.,
44 Sibbald,P.R. and Argos,P. (1990) J. Mol. Biol., 216, 813–818. Gnoj,L., Gottesman,T., Granat,S., Hameed,A., Kaplan,N., et al. (1997)
45 Sander,C. and Schneider,R. (1991) Proteins, 9, 56–68. GenBank accession no. 2104545.
46 Gerstein,M., Sonnhammer,E.L. and Chothia,C. (1994) J. Mol. Biol., 236, 73 Nagase,T., Seki,N., Ishikawa,K., Ohira,M., Kawarabayasi,Y., Ohara,O.,
1067–1078. Tanaka,A., Kotani,H., Miyajima,N. and Nomura,N. (1996) DNA Res., 3,
47 Henikoff,S. and Henikoff,J.G. (1994) J. Mol. Biol., 243, 574–578. 321–329.
48 Thompson,J.D., Higgins,D.G. and Gibson,T.J. (1994) Comput. Appl. 74 Wilson,R., Ainscough,R., Anderson,K., Baynes,C., Berks,M., Bonfield,J.,
Biosci., 10, 19–29. Burton,J., Connell,M., Copsey,T., Cooper,J., et al. (1994) Nature, 368, 32–38.
49 Eddy,S.R., Mitchison,G. and Durbin,R. (1995) J. Comput. Biol., 2, 9–23. 75 Barrell,B.G., Rajandream,M.A. and Connor,R.E. (1996) EMBL accession
50 Gotoh,O. (1995) Comput. Appl. Biosci., 11, 543–551. no. Z81317.
51 Krogh,A. and Mitchison,G. (1995) In Rawlings,C., Clark,D., Altman,R., 76 Johnston,M., Andrews,S., Brinkman,R., Cooper,J., Ding,H., Du,Z.,
Hunter,L., Lengauer,T. and Wodak,S. (eds.), Proceedings of the Third Favello,A., Fulton,L., Gattung,S., Greco,T., et al. (1994) GenBank
International Conference on Intelligent Systems for Molecular Biology. accession no. 662142.
AAAI Press, Menlo Park, CA, pp. 215–221. 77 Kaneko,T., Sato,S., Kotani,H., Tanaka,A., Asamizu,E., Nakamura,Y.,
52 Henikoff,J.G. and Henikoff,S. (1996) Comput. Appl. Biosci., 12, 135–143. Miyajima,N., Hirosawa,M., Sugiura,M., Sasamoto,S., et al. (1996) DNA
53 Brown,M., Hughey,R., Krogh,A., Mian,I.S., Sjölander,K. and Haussler,D. Res., 3, 109–136.
(1993) In Hunter,L., Searls,D. and Shavlik,J. (eds.), Proceedings of the 78 Tabata,S. (1996) GenBank accession no. 1651660.
First International Conference on Intelligent Systems for Molecular 79 Allende,M.L., Amsterdam,A., Becker,T., Kawakami,K., Gaiano,N. and
Biology. AAAI Press, Menlo Park, CA, pp. 47–55. Hopkins,N. (1996) Genes Dev., 10, 3141–3155.
80 Hernandez,K., Weber,N., Wipfli,P. and Schmidheini,T. (1996) EMBL
54 Bailey,T.L. and Gribskov,M. (1996) In States,D.J., Agarwal,P.,
accession no. Z72888.
Gaasterland,T., Hunter,L. and Smith,R. (eds.), Proceedings of the Fourth
81 Sonnhammer,E.L. and Durbin,R. (1994) Comput. Appl. Biosci., 10,
International Conference on Intelligent Systems for Molecular Biology.
301–307.
AAAI Press, Menlo Park, CA, pp. 15–24. 82 Gotoh,O. (1982) J. Mol. Biol., 162, 705–708.
55 Karplus,K. (1995) In Rawlings,C., Clark,D., Altman,R., Hunter,L., 83 Fitch,W.M. and Smith,T.F. (1983) Proc. Natl. Acad. Sci. USA, 80,
Lengauer,T. and Wodak,S. (eds.), Proceedings of the Third International 1382–1386.
Conference on Intelligent Systems for Molecular Biology. AAAI Press, 84 Altschul,S.F. and Erickson,B.W. (1986) Bull. Math. Biol., 48, 603–616.
Menlo Park, CA, pp. 188–196. 85 Myers,E.W. and Miller,W. (1988) Comput. Appl. Biosci., 4, 11–17.
56 Sjölander,K., Karplus,K., Brown,M., Hughey,R., Krogh,A., Mian,I.S. and 86 Altschul,S.F. Proteins, in press.
Haussler,D. (1996) Comput. Appl. Biosci., 12, 327–345. 87 Richardson,M., Dilworth,M.J. and Scawen,M.D. (1975) FEBS Lett., 51,
57 Wootton,J.C. and Federhen,S. (1993) Comput. Chem., 17, 149–163. 33–37.
58 Altschul,S.F., Boguski,M.S., Gish,W. and Wootton,J.C. (1994) Nature 88 Matsuda,G., Maita,T., Braunitzer,G. and Schrank,B. (1980) Hoppe-Seyler’s
Genet., 6, 119–129. Z. Physiol. Chem., 361, 1107–1116.
59 Benson,D.A., Boguski,M.S., Lipman,D.J. and Ostell,J. (1997) Nucleic 89 Tokunaga,O., Yaegashi,T., Lowe,J., Dobbs,L. and Padmanabhan,R. (1986)
Acids Res., 25, 1–6. Virology, 155, 418–433.
60 Holm,L. and Sander,C. (1997) Structure, 5, 165–171. 90 Fulton,L. and Waterston,R. (1995) GenBank accession no. 1176713.