BLAST*

Table of Contents
FUNCTION
DESCRIPTION
OUTPUT
INTERPRETING OUTPUT
INPUT FILES
RELATED PROGRAMS
RESTRICTIONS
CHOOSING SEARCH SETS
ALGORITHM
Preliminaries
Turning Hits Into HSPs
Generating Gapped Extensions
CONSIDERATIONS
SUGGESTIONS
FILTERING OUT LOW COMPLEXITY SEQUENCES
AMINO ACID SCORING
NUCLEOTIDE SCORING
ALTERNATIVE GENETIC CODES
NETWORK CONSIDERATIONS
CITING BLAST
ACKNOWLEDGEMENT
PARAMETER REFERENCE

FUNCTION

[ Top | Next ]

BLAST searches one or more nucleic acid or protein databases for sequences similar to one or more query sequences of any type. BLAST can produce gapped alignments for the matches it finds.

DESCRIPTION

[ Previous | Top | Next ]

BLAST, or Basic Local Alignment Search Tool, uses the method of Altschul et al. (J. Mol. Biol. 215: 403-410 (1990)) to search for similarities between a query sequence and all the sequences in a database.

This release of BLAST implements version 2 of BLAST from the National Center for Biotechnology Information (NCBI) described in Altschul et al. (Nucleic Acids Res. 25(17): 3389-3402 (1997)). BLAST is known as "gapped BLAST" because, in addition to offering a three-fold speedup over the original BLAST, it generates gapped alignments between query and database sequences.

You can specify any number of query sequences to BLAST, and they may be in any combination of protein or nucleic acid sequences. You can also specify any number of databases to BLAST, as long as all of the databases are of the same type (protein or nucleic acid). In the current release, if you want to specify multiple databases you must do so on the command line, for example

% blast -INfile2=PIR,SWPLUS

You cannot specify more than one database from the interactive menu. The query sequences and database(s) you search can be any combination of protein or nucleic acid.

The GCG BLAST program supports five different programs in the BLAST family:

BLASTP, Protein Query Searching a Protein Database

Each database sequence is compared to each query in a separate protein-protein pairwise comparison.

BLASTX, Nucleotide Query Searching a Protein Database

Each query is translated, and each of the six products is compared to each database sequence in a separate protein-protein pairwise comparison.

BLASTN, Nucleotide Query Searching a Nucleotide Database

Each database sequence is compared to the query in a separate nucleotide-nucleotide pairwise comparison.

TBLASTN, Protein Query Searching a Nucleotide Database

Each nucleotide database sequence is translated, and each of the six products is compared to the queries in a separate protein-protein pairwise comparison.

TBLASTX, Nucleotide Query Searching a Nucleotide Database

The query and database sequences are translated in six frames, and each of the 12 products (for each query sequence) is compared in 36 different pairwise comparisons. Because this program involves more computation than the others, gapped alignments are not available when using TBLASTX.

Normally, BLAST decides which BLAST program you want to use simply by looking at the type (protein or nucleic acid) of your query sequence and the database you have selected. In the case of nucleotide-nucleotide searches, there are two programs that can do the search. By default, BLASTN is used. To search using TBLASTX instead, use -TBLASTX (but remember that gapped alignments are not available when using TBLASTX).

BLAST performs only local searches: it searches databases maintained at your institution. Local searches can consume significant computing resources, and require diligent maintenance of local databases. If you would rather run BLAST searches remotely, use NetBLAST. Keep in mind, however, that using NetBLAST is a security risk if your sequence data are confidential.

BLAST is a statistically driven search method that finds regions of similarity between your query and database sequences and produces gapped alignments of these regions. Within these aligned regions, the sum of the scoring matrix values of their constituent symbol pairs is higher than some level that you would expect to occur by chance alone.

You are prompted to set an expectation level for the entire search. By default this level is 10.0, which means that hits are reported only if they have a score that would be expected to occur purely by chance nore more than 10 times in this particular search.

OUTPUT

[ Previous | Top | Next ]

Below is part of the output from the search in the example session:


///////////////////////////////////////////////////////////////////////////////

BLASTP 2.0.5 [May-5-1998]

Reference: Altschul, Stephen F., Thomas L. Madden, Alejandro A. Schaffer,
Jinghui Zhang, Zheng Zhang, Webb Miller, and David J. Lipman (1997),
"Gapped BLAST and PSI-BLAST: a new generation of protein database search
programs",  Nucleic Acids Res. 25:3389-3402.
Query= PIR1:MYWHP
         (153 letters)

Database: pir
           107,076 sequences; 34,138,851 total letters

Searching. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
///////////////////////////////////////////////////////////////////////////////
                                                                   Score     E
Sequences producing significant alignments:                        (bits)  Value
 ..
PIR1:MYWHP  Begin: 1 End: 153 !myoglobin - sperm whale                313  1e-85
PIR1:MYWHW  Begin: 1 End: 153 !myoglobin - dwarf sperm whale          305  3e-83
///////////////////////////////////////////////////////////////////////////////

PIR1:MYSLG  Begin: 2 End: 153 !myoglobin - gray seal                  270  1e-72
///////////////////////////////////////////////////////////////////////////////

PIR1:MYAQ  Begin: 3 End: 154 !myoglobin - American alligator          206  2e-53
\\End of List
>PIR1:MYWHP myoglobin - sperm whale
           Length = 153

 Score =  313 bits (794), Expect = 1e-85
 Identities = 153/153 (100%), Positives = 153/153 (100%)

Query: 1   VLSEGEWQLVLHVWAKVEADVAGHGQDILIRLFKSHPETLEKFDRFKHLKTEAEMKASED 60
           VLSEGEWQLVLHVWAKVEADVAGHGQDILIRLFKSHPETLEKFDRFKHLKTEAEMKASED
Sbjct: 1   VLSEGEWQLVLHVWAKVEADVAGHGQDILIRLFKSHPETLEKFDRFKHLKTEAEMKASED 60

Query: 61  LKKHGVTVLTALGAILKKKGHHEAELKPLAQSHATKHKIPIKYLEFISEAIIHVLHSRHP 120
           LKKHGVTVLTALGAILKKKGHHEAELKPLAQSHATKHKIPIKYLEFISEAIIHVLHSRHP
Sbjct: 61  LKKHGVTVLTALGAILKKKGHHEAELKPLAQSHATKHKIPIKYLEFISEAIIHVLHSRHP 120

Query: 121 GDFGADAQGAMNKALELFRKDIAAKYKELGYQG 153
           GDFGADAQGAMNKALELFRKDIAAKYKELGYQG
Sbjct: 121 GDFGADAQGAMNKALELFRKDIAAKYKELGYQG 153

>PIR1:MYWHW myoglobin - dwarf sperm whale
           Length = 153

 Score =  305 bits (773), Expect = 3e-83
 Identities = 148/153 (96%), Positives = 151/153 (97%)

Query: 1   VLSEGEWQLVLHVWAKVEADVAGHGQDILIRLFKSHPETLEKFDRFKHLKTEAEMKASED 60
           VLSEGEWQLVLHVWAKVEAD+AGHGQDILIRLFK HPETLEKFDRFKHLK+EAEMKASED
Sbjct: 1   VLSEGEWQLVLHVWAKVEADIAGHGQDILIRLFKHHPETLEKFDRFKHLKSEAEMKASED 60

Query: 61  LKKHGVTVLTALGAILKKKGHHEAELKPLAQSHATKHKIPIKYLEFISEAIIHVLHSRHP 120
           LKKHGVTVLTALGAILKKKGHHEAELKPLAQSHATKHKIPIKYLEFISEAIIHVLHSRHP
Sbjct: 61  LKKHGVTVLTALGAILKKKGHHEAELKPLAQSHATKHKIPIKYLEFISEAIIHVLHSRHP 120

Query: 121 GDFGADAQGAMNKALELFRKDIAAKYKELGYQG 153
            DFGADAQGAM+KALELFRKDIAAKYKELGYQG
Sbjct: 121 ADFGADAQGAMSKALELFRKDIAAKYKELGYQG 153

///////////////////////////////////////////////////////////////////////////////
>PIR1:MYSLG myoglobin - gray seal
           Length = 153

 Score =  270 bits (682), Expect = 1e-72
 Identities = 127/152 (83%), Positives = 140/152 (91%)

Query: 2   LSEGEWQLVLHVWAKVEADVAGHGQDILIRLFKSHPETLEKFDRFKHLKTEAEMKASEDL 61
           LS+GEW LVL+VW KVE D+AGHGQ++LIRLFKSHPETLEKFD+FKHLK+E +M+ SEDL
Sbjct: 2   LSDGEWHLVLNVWGKVETDLAGHGQEVLIRLFKSHPETLEKFDKFKHLKSEDDMRRSEDL 61

Query: 62  KKHGVTVLTALGAILKKKGHHEAELKPLAQSHATKHKIPIKYLEFISEAIIHVLHSRHPG 121
           +KHG TVLTALG ILKKKGHHEAELKPLAQSHATKHKIPIKYLEFISEAIIHVLHS+HP
Sbjct: 62  RKHGNTVLTALGGILKKKGHHEAELKPLAQSHATKHKIPIKYLEFISEAIIHVLHSKHPA 121

Query: 122 DFGADAQGAMNKALELFRKDIAAKYKELGYQG 153
           +FGADAQ AM KALELFR DIAAKYKELG+ G
Sbjct: 122 EFGADAQAAMKKALELFRNDIAAKYKELGFHG 153

///////////////////////////////////////////////////////////////////////////////
>PIR1:MYAQ myoglobin - American alligator
           Length = 154

 Score =  206 bits (519), Expect = 2e-53
 Identities = 94/152 (61%), Positives = 122/152 (79%)

Query: 2   LSEGEWQLVLHVWAKVEADVAGHGQDILIRLFKSHPETLEKFDRFKHLKTEAEMKASEDL 61
           LS+ EW+ VL +W KVE+ +  HG +++IRL + HPET E+F++FKH+KT  EMK+SE +
Sbjct: 3   LSDQEWKHVLDIWTKVESKLPEHGHEVIIRLLQEHPETQERFEKFKHMKTADEMKSSEKM 62

Query: 62  KKHGVTVLTALGAILKKKGHHEAELKPLAQSHATKHKIPIKYLEFISEAIIHVLHSRHPG 121
           K+HG TV TALG ILK+KG+H   LKPLA+SHA +HKIP+KYLEFISE I+ V+  ++P
Sbjct: 63  KQHGNTVFTALGNILKQKGNHAEVLKPLAKSHALEHKIPVKYLEFISEIIVKVIAEKYPA 122

Query: 122 DFGADAQGAMNKALELFRKDIAAKYKELGYQG 153
           DFGAD+Q AM KALELFR D+A+KYKE GYQG
Sbjct: 123 DFGADSQAAMRKALELFRNDMASKYKEFGYQG 154

///////////////////////////////////////////////////////////////////////////////
  Database: pir
    Posted date:  Jul 8, 1998  4:54 PM
  Number of letters in database: 34,138,851
  Number of sequences in database:  107,076

Lambda     K      H
   0.318    0.135    0.392

Gapped
Lambda     K      H
   0.270   0.0470    0.230

Matrix: BLOSUM62
Gap Penalties: Existence: 11, Extension: 1
Number of Hits to DB: 10131438
Number of Sequences: 107076
Number of extensions: 379101
Number of successful extensions: 1983
Number of sequences better than 10: 645
Number of HSP's better than 10.0 without gapping: 572
Number of HSP's successfully gapped in prelim test: 73
Number of HSP's that attempted gapping in prelim test: 1152
Number of HSP's gapped (non-prelim): 652
length of query: 153
length of database: 34138851
effective HSP length: 50
effective length of query: 103
effective length of database: 28785051
effective search space: 2964860253
T: 11
A: 40
X1: 16 ( 7.3 bits)
X2: 38 (14.8 bits)
X3: 64 (24.9 bits)
S1: 41 (21.7 bits)
S2: 61 (28.2 bits)

The output has four parts: 1) an introduction that tells where the search occurred and what database and query were compared; 2) a list of the sequences in the database containing HSPs (high-scoring segment pairs) whose scores were least likely to have occurred by chance (the entries in this list have begin and end ranges on them if -FRAGments is specified); 3) a display of the alignments of the HSPs showing identical and similar residues; and 4) a complete list of the parameter settings used for the search.

By default, BLAST looks for alignments that contain gaps. If you only look for alignments that do not contain gaps, there will often be more than one segment pair associated with each database sequence.

The BLAST output is a list file that is suitable for input to any GCG program that allows indirect file specifications. For information about indirect file specification, see Chapter 2 of the User's Guide, Using Sequence Files and Databases.

INTERPRETING OUTPUT

[ Previous | Top | Next ]

Bit Score

Each aligned segment pair has a normalized score expressed in bits that lets you estimate the magnitude of the search space you would have to look through before you would expect to find an HSP score as good as or better than this one by chance. If the bit score is 30, you would have to score, on average, about 1 billion independent segment pairs (2(30)) to find a score this good by chance. Each additional bit doubles the size of the search space. This bit score represents a probability; one over two raised to this power is the probability of finding such a segment by chance. Bit scores represent a probability level for sequence comparisons that is independent of the size of the search.

The size of the search space is proportional to the product of the query sequence length times the sum of the lengths of the sequences in the database. This product, referred to as N in Altschul's publications, is multiplied by a coefficient K to get the size of the search space. When searching protein databases with protein queries, K is about 0.13. BLAST uses estimates of K produced before it runs by random simulation (Altschul & Gish, Methods in Enzymology 266; 460-480 (1996)).

E Value

There is a probability associated with each pairwise comparison in the list and with each segment pair alignment. The number shown in the list is the probability that you would observe a score or group of scores as high as the observed score purely by chance when you do a search against a database of this size.

An ideal search would find hits that go from extremely unlikely to ones whose best scores should have occurred by chance alone (that is, with probabilities approaching 1.0).

N

If you specify ungapped alignments to BLAST, a third column of data will appear in your output under the heading N. The number in that column indicates how many HSPs were involved in computing the statistics for the sequence. If the number is greater than 1, the scores of multiple HSPs were combined to produce the result. See the ALGORITHM topic for more information.

BLAST Parameters

At the end of the output is a listing of parameter settings along with some trace information about the search. Some of these parameters are described in this document, but to get complete documentation on these parameters, look at the BLAST release notes on the World Wide Web at

http://www.ncbi.nlm.nih.gov/BLAST/newblast.html.

INPUT FILES

[ Previous | Top | Next ]

BLAST accepts any number of protein or nucleic acid sequences as input. The search set is a specially formatted database. See the GCGToBLAST entry in the Program Manual for information on how to create a local database that BLAST can search from a set of sequences in GCG format.

The function of BLAST depends on whether your input sequence(s) are protein or nucleotide. Programs determine the type of a sequence by the presence of either Type: N or Type: P on the last line of the text heading just above the sequence. If your sequence(s) are not the correct type, turn to Appendix VI for information on how to change or set the type of a sequence.

RELATED PROGRAMS

[ Previous | Top | Next ]

NetBLAST searches for sequences similar to a query sequence. The query and the database searched can be either peptide or nucleic acid in any combination. NetBLAST can search only databases maintained at the National Center for Biotechnology Information (NCBI) in Bethesda, Maryland, USA.

GCGToBLAST combines any set of GCG sequences into a database that you can search with BLAST.

FastA does a Pearson and Lipman search for similarity between a query sequence and a group of sequences of the same type (nucleic acid or protein). For nucleotide searches, FastA may be more sensitive than BLAST. TFastA does a Pearson and Lipman search for similarity between a protein query sequence and any group of nucleotide sequences. TFastA translates the nucleotide sequences in all six reading frames before performing the comparison. It is designed to answer the question, "What implied protein sequences in a nucleotide sequence database are similar to my protein sequence?" FastX does a Pearson and Lipman search for similarity between a nucleotide query sequence and a group of protein sequences, taking frameshifts into account. FastX translates both strands of the nucleic sequence before performing the comparison. It is designed to answer the question, "What implied protein sequences in my nucleic acid sequence are similar to sequences in a protein database?" TFastX does a Pearson and Lipman search for similarity between a protein query sequence and any group of nucleotide sequences, taking frameshifts into account. It is designed to be a replacement for TFastA, and like TFastA, it is designed to answer the question, "What implied protein sequences in a nucleotide sequence database are similar to my protein sequence?"

SSearch does a rigorous Smith-Waterman search for similarity between a query sequence and a group of sequences of the same type (nucleic acid or protein). This may be the most sensitive method available for similarity searches. Compared to BLAST and FastA, it can be very slow.

ProfileSearch uses a profile (representing a group of aligned sequences) as a query to search the database for new sequences with similarity to the group. The profile is created with the program ProfileMake. MotifSearch uses a set of profiles (representing similarities within a family of sequences) as a query to either a) search a database for new sequences similar to the original family, or b) annotate the members of the the original family with details of the matches between the profiles and each of the members. Normally, the profiles are created with the program MEME. FrameSearch searches a group of protein sequences for similarity to one or more nucleotide query sequences, or searches a group of nucleotide sequences for similarity to one or more protein query sequences. For each sequence comparison, the program finds an optimal alignment between the protein sequence and all possible codons on each strand of the nucleotide sequence. Optimal alignments may include reading frame shifts.

FindPatterns, StringSearch, and Names are other sequence identification programs.

RESTRICTIONS

[ Previous | Top | Next ]

Because of the way BLAST must estimate certain statistical parameters (see the ALGORITHM topic elsewhere in this document), the number of scoring matrices available for use with BLAST is limited. Currently, valid choices for the Protein scoring matrix parameter are BLOSUM62 (the default), BLOSUM45, BLOSUM80, PAM30, and PAM70.

Gap creation and gap extension penalties are supported in limited combinations depending upon which scoring matrix is in use. The following table shows the allowed combinations for amino acids. The first values listed are the defaults for each scoring matrix.





Scoring Matrix Gap Opening Penalty Gap Extension Penalty




BLOSUM62 11 1

7 2
8 2
9 2
10 1
12 1



BLOSUM80 10 1

6 2
7 2
8 2
9 1
11 1



BLOSUM45 14 2

10 3
11 3
12 3
13 3
12 2
13 2
15 2
16 1
17 1
18 1
19 1



PAM30 9 1

5 2
6 2
7 2
8 1
10 1



PAM70 10 1

6 2
7 2
8 2
9 1
11 1




Gapped alignments are not an option when running TBLASTX.

You may choose multiple query sequences, any of which may be either nucleic acid or protein. You may also choose multiple databases against which to search, however each of these must be of the same type.

If you used GCGToBLAST to create your BLAST databases from any source other than a GCG-formatted database (such as from an arbitrary sequence file, an MSF or RSF file, etc.), then BLAST's list file output won't be a functional list file. If you want to take full advantage of BLAST's list file output, make sure that you generate your BLAST databases from a GCG-formatted database. You can use DataSet to generate such databases from any set of sequences in GCG format.

CHOOSING SEARCH SETS

[ Previous | Top | Next ]

BLAST can search only a specially compressed form of the data. Therefore, you can search only those databases that are available in this form, and you must search them in their entirety. If you want to restrict the search to a specific set of sequences, use the program GCGToBLAST to create a specially compressed database consisting of just those sequences.

ALGORITHM

[ Previous | Top | Next ]

BLAST is a client for an implementation of gapped BLAST (Altschul et al., Nucleic Acids Research 25; 3389-3402 (1997)), an heuristic algorithm for searching protein and nucleic acid databases for similarities to query sequences.

In this discussion, we will use BLASTP, which searches for similarities between protein queries and protein databases, as a prototype for BLAST. However, the ideas are immediately applicable to comparisons involving conceptual translations of query sequences and databases, and extend to similarity searches between nucleic acid sequences as well.

To describe the algorithm briefly, BLAST compares a query sequence with a database sequence by first locating two non-overlapping sequence segments in common within a certain distance of each other, and then attempting to extend these so-called "hits" into locally optimal alignments between the sequences being compared. A more detailed description is provided below.

Preliminaries

[ Previous | Top | Next ]

BLAST uses a substitution matrix (such as the BLOSUM or PAM matrices) to assign a score to the alignment of any pair of amino acids. An aggregate score for an alignment segment can be computed by summing the scores of each amino acid pair in that segment. When given two sequences to compare, the original (ungapped) BLAST algorithm searches for arbitrary but equal length segments within each sequence that have a maximal aggregate score which meets or exceeds some threshold or cutoff score. BLAST looks for locally optimal alignments between the two sequences whose scores cannot be improved either by extending or trimming. Such locally optimal alignments are called "high-scoring segment pairs," or HSPs.

If we assume a simple protein model in which amino acids occur randomly at all positions and in proportion to the frequencies at which they are found within the database and query sequences, then we can compute a normalized score (expressed in units called bits) from the nominal score of an HSP. Such normalized scores allow direct statistical comparison of results regardless of the scoring system used (see "Generating Gapped Extensions" for a caveat to this). Furthermore, the normalized score can be used to compute an expect value, or E-value, which is the number of distinct HSPs having at least that normalized score expected to occur by chance. This theory has not been proved for gapped local alignments and their associated scores, but there are indications that it remains valid (Altschul et al., 1997).

Turning Hits Into HSPs

[ Previous | Top | Next ]

The central idea of the BLAST algorithm is that any statistically significant alignment between two sequences is likely to contain a high-scoring pair of aligned words. A word is simply a sequence segment of specified length (usually 3 for protein sequences). BLAST begins its comparison of a query sequence to a database by scanning the database for words that score at least the threshold score T when aligned with some word within the query sequence. Any word pair satisfying this condition is called a hit. The diagonal of a hit involving words starting at positions (x, y) of the database and query sequences is defined as x-y. The distance between two hits on the same diagonal is defined as the difference between their first coordinates.

Once a hit is found, BLAST determines whether the hit lies within an alignment having an aggregate score high enough to be reported. It does this by extending the hit in both directions until the running alignment's score has dropped more than some quantity X below the maximum score yet attained. This extension step is quite costly, taking upwards of 90% of BLAST's execution time under most circumstances.

In order to reduce the number of extensions it has to perform, BLAST takes advantage of the fact that an interesting HSP is typically much longer than a single hit. In fact, it is likely to contain multiple hits on the same diagonal within a relatively short distance of one another. Therefore, BLAST chooses a length A and invokes an ungapped extension if and only if two non-overlapping hits are found on the same diagonal within distance A of one another (any hit that overlaps the most recent one is ignored).

Generating Gapped Extensions

[ Previous | Top | Next ]

Gapped extensions allow BLAST to maintain its sensitivity while tolerating a much higher chance of missing any single moderately scoring HSP. However, gapped extensions take about 500 times longer to execute than ungapped extensions. Therefore, BLAST triggers a gapped extension for an HSP only when its score exceeds a moderate score Sg specifically chosen so that no more than about one gapped extension is invoked per 50 database sequences.

To generate the gapped local alignment, BLAST uses a standard dynamic programming algorithm for pairwise sequence alignment which traverses the cells of a path graph, the dimensions of which are the lengths of the two sequences being compared, performing a fixed amount of computation per each cell. Starting from a single aligned pair of residues, called the seed, the dynamic programming proceeds both forward and backward through the path graph considering only those cells for which the optimal local alignment score falls no more than X below the best alignmnet score yet found (this is a simple generalization of BLAST's method for constructing HSPs). The region of the path graph explored adapts to the alignment being produced.

The seed for the dynamic programming is the central residue pair of the length-11 segment of the HSP having the highest alignment score. If the HSP itself is shorter than 11 residues in length, its central pair of residues is chosen.

The resulting gapped alignment is reported only if it has an E-value low enough to be of interest. For any alignment actually reported, BLAST performs a gapped extension that records "traceback" information (Sankoff and Kruskal, 1983) using a substantially larger X parameter than that employed during the search stage to increase the accuracy of the alignment.

Because BLAST produces gapped alignments only for those few database sequences likely to be related to the query, it cannot estimate the parameters necessary to compute normalized scores on the fly. Instead, BLAST must rely on estimates of these parameters generated beforehand by random simultion. For this reason, BLAST cannot use a scoring system for which no simulation has been performed and still produce accurate estimates of statistical significance.

CONSIDERATIONS

[ Previous | Top | Next ]

Bit Scores and the Size of the Search

Altschul has shown that for sequences that have diverged by a certain amount, there is an informativeness (or ability to discriminate between chance scores and significant scores) associated with each residue pair in the segment pair. This informativeness is the amount of information obtainable from each residue pair in a real alignment that can be used to distinguish the real alignment from a random one. This informativeness can be expressed in bits. The sum of the information available from each residue pair in a segment is the segment pair's score in bits. Such scores are intuitively understandable as the significance of a segment pair score. To express such scores as a fraction you would divide 1 by 2 to the number of bits in the score. For example, if a segment pair has a bit-score of 16, then the appropriate fraction (1/2(16)=1/65,536) would suggest that you should see a score this high by chance about once for every 65,000 independent segment pairs you examine.

For nucleotide sequences that have not diverged, there should be an informativeness of about 2 bits per nucleotide pair. For protein sequences that have not diverged, the informativeness should be slightly over 4 bits per amino acid pair. (The informativeness per pair goes down as the sequences diverge and a segment pair score is maximally informative only when a scoring matrix appropriate to the extent of divergence between the sequences is used to calculate the score.)

The bit scores are absolute, but the expectation of finding any particular score depends on the size of the search space. The number of places where a segment pair might originate is proportional to the product of the length of the query times the sum of the lengths of all the sequences searched. This product is multiplied by a coefficient K to get the size of the search space. When searching protein databases with protein queries, K is approximately 0.13.

For a query sequence of length 300 aa searching a database of 12 million residues, the size of the search space would be 300 x 12,000,000 x 0.13 or 468,000,000. For a search this size, a score that only occurs once in every 65,000 potential segment pairs (that is, with a bit score of 16) would be expected to occur about 7,200 times by chance alone.

If the database being searched is highly redundant (as it might be if it contained several hundred homologous cytochromes), then size of the search space calculated by these methods will overestimate the size of the real search space.

Using BLAST for Nucleotide Searches

The detection of distant relationships between proteins is easier than between nucleotide sequences, even if the nucleotide sequences have to be translated in all six frames to make the amino acid comparison. To give a rough magnitude to this generalization, it is possible to detect similarities in proteins that have diverged by 250 substitutions per 100 residues (250 PAM units) while nucleotide similarities become obscure at distances much greater than 50 substitutions per 100 nucleotides (50 DNA PAM units). Nonetheless, when the nucleotide sequences being compared do not code for proteins, you have no alternative but to search at the nucleotide level. We suggest you consider either reducing the word size for BLAST from its default of 11 to perhaps six or seven, or using the FastA program when looking for nucleotide homologs.

Increasing Program Speed Using Multithreading

This program is multithreaded. It has the potential to run faster on a machine equipped with multiple processors because different parts of the analysis can be run in parallel on different processors. By default, the program assumes you have one processor, so the analysis is performed using one thread. You can use Number of processors to use for the search to increase the number of threads up to the number of physical processors on the computer.

Under ideal conditions, the increase in speed is roughly linear with the number of processors used. But conditions are rarely ideal. If your computer is heavily used, competition for the processors can reduce the program's performance. In such an environment, try to run multithreaded programs during times when the load on the system is light.

As the number of threads increases, the amount of memory required increases substantially. You may need to ask your system administrator to increase the memory quota for your account if you want to use more than two threads.

Never use Number of processors to use for the search to set the number of threads higher than the number of physical processors that the machine has -- it does not increase program performance, but instead uses up a lot of memory needlessly and makes it harder for other users on the system to get processor time. Ask your system administrator how many processors your computer has if you aren't sure.

When Blastall Produces No Output

You may see an error indicating that blastall produced no output (blastall is the name of the BLAST executable provided by NCBI). One of the possible causes of this condition is the presence of a file in your home directory called ".ncbirc" (under Unix) or "ncbi.cfg" (under VMS) which contains an invalid path to the NCBI data directory. The NCBI data directory should contain seqcode.val, gc.code, BLOSUM62, and perhaps some other data files. If your home directory does indeed contain such a file, we recommend that you either rename it (the safest option), edit it to update the path to the NCBI data directory (this takes some effort, but that path is contained in the logical name "NCBI"), or delete it (the simplest option). Your system administrator should be able to help you do this if you have trouble, or you may call GCG's techinical support staff at (608) 231-5200.

SUGGESTIONS

[ Previous | Top | Next ]

List Size Limit

A list size that is too small to display all the significant hits is a common problem. To see the unlisted hits you must run the search again with the list size limit set high enough to include everything significant.

Segment Pair Alignment Limit

BLAST displays alignments of segment pairs from the top 250 sequences in the list. You can adjust this limit with -ALIgnments. BLAST will not show alignments for sequences not present in the list.

Sensitivity

For nucleotide sequence comparisons, the word size defaults to 11 -- no segment pair can be scored unless it contains a perfect match of at least 11 consecutive bases. If sensitivity is much more important than selectivity, and your search cannot be done at the amino acid level, you might want to reduce the word size to seven or even six. NCBI has stated that there is only a marginal increase in sensitivity for settings smaller than this.

BLAST uses a word size of three for proteins (11 for blastn searches), which is appropriate for a wide range of searches, but you can adjust the synonym threshold T downwards to increase sensitivity at the price of speed. Read the PARAMETER REFERENCE topic for more information on -HITEXTHRESHold and -EXPect.

Batch Queue

Using BLAST to search a large local database can take a long time. You may want to run searches in the batch queue. You can specify that this program run at a later time in the batch queue by using -BATch. Run this way, the program prompts you for all the required parameters and then automatically submits itself to the batch or at queue. For more information, see "Using the Batch Queue" in Chapter 3, Using Programs in the User's Guide.

Relationship to FastA

For protein database searches, BLAST and FastA have similar sensitivity, although the different algorithms employed make it possible, at least in principle, for FastA to find things that BLAST misses and vice versa. For nucleotide database searches with nucleotide query sequences, FastA may be more sensitive, since by default BLAST ignores segment pairs that do not contain a perfect match of at least 11 adjacent nucleotides (22 bits). This default misses many obviously significant relationships. If you are looking for nucleotide sequence homologs that do not code for proteins (that is, if your search cannot be done at the amino acid level), we suggest you either reduce the word size to seven or use the FastA program instead of BLAST.

FILTERING OUT LOW COMPLEXITY SEQUENCES

[ Previous | Top | Next ]

BLAST filters out regions of low complexity from query sequences by default. You can turn filtering off by using the -NOFILter parameter. Searches against a nucleotide database with nucleotide queries (blastn) employ the DUST filter program (Hancock and Armstrong, Comput. Appl. Biosci. 10: 67-70 (1994); Tatusov and Lipman, in preparation). All other searches employ the SEG filter program (Wootton and Federhen, Computers in Chemistry 17: 149-163 (1993); Wootton and Federhen, Methods in Enzymology 266: 554-571 (1996)). For a general discussion of the role of filtering in search strategies, see Altschul et al., Nature Genetics 6: 119-129 (1994).

Short repeats and low complexity sequences, such as glutamine-rich regions, confound most database searching methods. For BLAST, the random model against which the significance of segment pair scores is evaluated assumes that at each position, each residue has a probability of occurring which is proportional to its composition in the database as a whole. Low complexity or highly repetitive sequences are inconsistent with this assumption.

Low complexity sequence found by the filter program is substituted using the letter N in nucleotide sequence and the letter X in amino acid sequence. Here is an example of a sequence aligned to a filtered copy of itself to show which parts are filtered out:

Low complexity sequence found by the filter program is substituted using the letter N in nucleotide sequence and the letter X in amino acid sequence. Here is an example of a sequence aligned to a filtered copy of itself to show which parts are filtered out:


  1 MAAKIFCLIMXXXXXXXXXXXXIFPQCSQAPIASLLPPYLSPAMSSVCENPILLPYRIQQ 60
  1 MAAKIFCLIMLLGLSASAATASIFPQCSQAPIASLLPPYLSPAMSSVCENPILLPYRIQQ 60

 61 AIAAGIXXXXXXXXXXXXXXXXXXXXXXXXXXNIRXXXXXXXXXXXXXXYSQQQQFLPFN 120
 61 AIAAGILPLSPLFLQQSSALLQQLPLVHLLAQNIRAQQLQQLVLANLAAYSQQQQFLPFN 120

121 QXXXXXXXXXXXXXXXXPFSQLAAAYPRQFLPFNQLAALNSHAYVXXXXXXPFSQLAAVS 180
121 QLAALNSAAYLQQQQLLPFSQLAAAYPRQFLPFNQLAALNSHAYVQQQQLLPFSQLAAVS 180

181 PAAFLTQQQLLPFYLHTAPNVGTXXXXXXXXXXXXXXXTNPAAFYQQPIIGGALF 235
181 PAAFLTQQQLLPFYLHTAPNVGTLLQLQQLLPFDQLALTNPAAFYQQPIIGGALF 235

AMINO ACID SCORING

[ Previous | Top | Next ]

BLAST normally uses the BLOSUM62 scoring matrix from Henikoff and Henikoff (Proc. Natl. Acad. Sci. USA 89; 10915-10919 (1992)) whenever the sequences being compared are proteins (including cases where nucleotide databases or query sequences are translated into protein sequences before comparison). You can use other BLOSUM45, BLOSUM80, or the more traditional PAM70 and PAM30 scoring matrices with Protein scoring matrix, for example Protein scoring matrix set to PAM40. Each matrix is most sensitive for finding homologs at the corresponding PAM distance. The seminal paper on this subject is Stephen Altschul's "Amino acid substitution matrices from an information theoretic perspective" (J. Mol. Biol. 219; 555-565 (1991)). If you are new to this literature, an easier place to start reading might be Altschul et al., "Issues in searching molecular sequence databases" (Nature Genetics, 6; 119-129 (1994)).

NUCLEOTIDE SCORING

[ Previous | Top | Next ]

There is no external scoring matrix for nucleotide-nucleotide searches (that is, searches where both the query and the database are nucleotide sequences and where you have not used -TBLASTX. But as is explained below, you can specify a nucleotide-nucleotide scoring matrix for any PAM distance by changing the match/mismatch ratio. The default ratio is +1/-3. You can change the ratio by specifying a new value for the numerator using -MATCH.

ALTERNATIVE GENETIC CODES

[ Previous | Top | Next ]

BLAST normally uses the standard genetic code if either the query or the database sequences requires translation. If your query comes from a system where this genetic code is inappropriate, you can select any of these alternative codes by the numbers given in the following table:


     1 Standard or Universal
     2 Vertebrate Mitochondrial
     3 Yeast Mitochondrial
     4 Mold, Protozoan, Coelenterate Mitochondrial and Mycoplasma/Spiroplasma
     5 Invertebrate Mitochondrial
     6 Ciliate Macronuclear
     7 [Do not use this index]
     8 [Do not use this index]
     9 Echinodermate Mitochondrial
    10 Alternative Ciliate(Euplotid) Macronuclear
    11 Eubacterial
    12 Alternative Yeast
    13 Ascidian Mitochondrial
    14 Flatworm Mitochondrial
    15 Alternate Ciliate (Blepharisma) Nuclear
    16 Chlorophycean Mitochondrial
    21 Trematode Mitochondrial

You can specify the genetic code for the query and the database independently. Use Genetic code for query translation set to Vertebrate Mitochondrial to tell BLAST to use the vertebrate mitochondrial code to translate the query. Use Genetic code for database translation set to Yeast Mitochondrial to tell BLAST to use the yeast mitochondrial code to translate the database. (Note that most of the genes in GenBank will be translated inappropriately if you select a nonstandard genetic code for database translation.)

NETWORK CONSIDERATIONS

[ Previous | Top | Next ]

BLAST searches only local databases. See the NetBlast entry in the Program Manual for information on how to run BLAST searches remotely.

CITING BLAST

[ Previous | Top | Next ]

The original paper describing BLAST is Altschul, Stephen F., Gish, Warren, Miller, Webb, Myers, Eugene W., and Lipman, David J. (1990). Basic local alignment search tool. J. Mol. Biol. 215; 403-410. Gapped BLAST is described in Altschul, Stephen F., Madden, Thomas L., Schaffer, Alejandro A., Zhang, Jinghui, Zhang, Zheng, Miller, Webb, and Lipman, David J. (1997). Gapped BLAST and PSI-BLAST: a new generation of protein database search programs. Nucleic Acids Res. 25(17); 3389-3402.

ACKNOWLEDGEMENT

[ Previous | Top | Next ]

BLAST was written by Warren Gish, formerly of the National Center for Biotechnology Information (NCBI), in collaboration with Stephen Altschul, Webb Miller, Eugene Myers, David Lipman, and David States. The document you are now reading was written by John Devereux, with some modifications for BLAST2 by Ted Slater.

Blastall (NCBI's implementation of BLAST 2.0) was written for NCBI by Tom Madden. GCG's client for blastall was written by Ted Slater for distribution with the Wisconsin Package Version 10.0; some portions were taken from GCG's original BLAST client written by Scott Rose. The output post-processor for release 10.0 was written by Ron Stewart.

We are extremely grateful to Stephen Altschul and Warren Gish for their careful and original work on BLAST and for their critical comments on GCG's BLAST documentation, and we are very grateful to NCBI for making these programs and services available to the molecular biology community.

PARAMETER REFERENCE

[ Previous | Top | Next ]

You can set the parameters listed below from the command line. For more information, see "Using Program Parameters" in Chapter 3, Using Programs in the User's Guide.

Following some of the optional parameters described below is a letter or short expression in parentheses. These are the names of the corresponding parameters at the bottom of your BLAST output. Where the parameter name has changed from the 1.x versions of BLAST from NCBI, the old parameter name is given as well.

Ignore hits that might occur more than how many times by chance alone

lets you influence the number of hits in your output by including scores that would be expected to have occurred by chance alone. There is nothing to prevent many biologically significant but statistically insignificant segment pairs from being screened out, so you may sometimes want to increase this parameter in order to have an opportunity to see them.

Maximum number of sequences listed in the output

use Maximum number of sequences listed in the output to limit the number of sequences in your output to any value between 0 (for blastall's program defaults) and 1000. The output may contain fewer sequences than the number you set here since only statistically significant sequences will be included.

Number of processors to use for the search

Use Number of processors to use for the search to use additional threads for the database search on a multiprocessor computer. Your system manager should limit the number of threads you can use to less than or equal to the number of processors the computer has available.

Word size

sets the size of the short regions of similarity between sequences that BLAST initially searches for. If Word size set to 0, BLAST uses the default values: 11 for blastn and 3 for the other programs. Smaller word sizes result in a more sensitive search at the expense of a longer search time.

Reward for all nucleotide matches

sets the reward for nucleotide matches (BLASTN only).

Penalty for all nucleotide mismatches

sets the penalty for nucleotide matches (BLASTN only).

Protein scoring matrix

sets the amino acid substitution matrix to use. BLAST normally uses the BLOSUM62 amino acid substitution matrix from Henikoff and Henikoff for protein sequence comparisons (including all cases where nucleotide database or query sequences are translated before comparison).

Gap creation penalty

sets the penalty for adding a gap to the alignment. See the RESTRICTIONS topic for more information about setting the gap opening penalty.

Gap extension penalty

sets the penalty for lengthening an existing gap in the alignment. See the RESTRICTIONS topic for more information about setting the gap extension penalty.

Filter input sequences for low complexity regions

suppresses filtering out low-complexity regions from query sequences.

By default, the SEG filter (Wootton and Federhen, Computers Chem. 17(2); 149-163 (1993)) masks low-complexity sequences in protein sequences, and the DUST filter (Hancock and Armstrong, Computer Applications in the Biosciences 10; 67-70 (1994)) performs the same function for nucleic acid sequences. Masked regions are excluded from the search. These regions are replaced with X's (in protein sequences) or N's (in nucleic acid sequences) in the output to let you identify the regions that were excluded.

Genetic code for query translation

    Standard or Universal
    Vertebrate Mitochondrial
    Yeast Mitochondrial
    Mold Mitochondrial and Mycoplasma
    Invertebrate Mitochondrial
    Ciliate Macronuclear
    Echinodermate Mitochondrial
    Alternative Ciliate Macronuclear
    Eubacterial
    Alternative Yeast
    Ascidian Mitochondrial
    Flatworm Mitochondrial
    Blepharisma (ciliate) nuclear
    Chlorophycean Mitochondrial
    Trematode Mitochondrial

sets a genetic code to use for the translation of the query sequence. BLAST uses the standard ("universal") genetic code unless you specify the number of one of the alternative codes listed under the topic ALTERNATIVE GENETIC CODES.

Genetic code for database translation

    Standard or Universal
    Vertebrate Mitochondrial
    Yeast Mitochondrial
    Mold Mitochondrial and Mycoplasma
    Invertebrate Mitochondrial
    Ciliate Macronuclear
    Echinodermate Mitochondrial
    Alternative Ciliate Macronuclear
    Eubacterial
    Alternative Yeast
    Ascidian Mitochondrial
    Flatworm Mitochondrial
    Blepharisma (ciliate) nuclear
    Chlorophycean Mitochondrial
    Trematode Mitochondrial

sets a genetic code to use for the translation of the database sequences. If you are searching for proteins from a system that doesn't use the standard ("universal") genetic code, you can select a more appropriate code from those listed under the topic ALTERNATIVE GENETIC CODES. Note that most of the genes in the nucleotide databases will be translated incorrectly if you select a nonstandard genetic code.

Create gapped alignments

performs non-gapped alignments. By default gapped alignments are performed except when using tblastx, where gapped alignments are not available.

Printed: February 15, 2000 6:48 (1162)