bwa.1 17.1 KB
Newer Older
.TH bwa 1 "16 October 2010" "bwa-0.5.8c" "Bioinformatics tools"
2 3 4 5 6 7 8 9 10 11 12 13 14
bwa - Burrows-Wheeler Alignment Tool
bwa index -a bwtsw database.fasta
bwa aln database.fasta short_read.fastq > aln_sa.sai
bwa samse database.fasta aln_sa.sai short_read.fastq > aln.sam
bwa sampe database.fasta aln_sa1.sai aln_sa2.sai read1.fq read2.fq > aln.sam
bwa bwasw database.fasta long_read.fastq > aln.sam
16 17 18 19 20 21 22 23 24 25

BWA is a fast light-weighted tool that aligns relatively short sequences
(queries) to a sequence database (targe), such as the human reference
genome. It implements two different algorithms, both based on
Burrows-Wheeler Transform (BWT). The first algorithm is designed for
short queries up to ~200bp with low error rate (<3%). It does gapped
global alignment w.r.t. queries, supports paired-end reads, and is one
of the fastest short read alignment algorithms to date while also
visiting suboptimal hits. The second algorithm, BWA-SW, is designed for
27 28
long reads with more errors. It performs heuristic Smith-Waterman-like
alignment to find high-scoring local hits (and thus chimera). On
low-error short queries, BWA-SW is slower and less accurate than the
30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163
first algorithm, but on long queries, it is better.
For both algorithms, the database file in the FASTA format must be
first indexed with the
.B `index'
command, which typically takes a few hours. The first algorithm is
implemented via the
.B `aln'
command, which finds the suffix array (SA) coordinates of good hits of
each individual read, and the
.B `samse/sampe'
command, which converts SA coordinates to chromosomal coordinate and
pairs reads (for `sampe'). The second algorithm is invoked by the
.B `dbtwsw'
command. It works for single-end reads only.

.B index
bwa index [-p prefix] [-a algoType] [-c] <in.db.fasta>

Index database sequences in the FASTA format.

.TP 10
.B -c
Build color-space index. The input fast should be in nucleotide space.
.B -p STR
Prefix of the output database [same as db filename]
.B -a STR
Algorithm for constructing BWT index. Available options are:
.B is
IS linear-time algorithm for constructing suffix array. It requires
5.37N memory where N is the size of the database. IS is moderately fast,
but does not work with database larger than 2GB. IS is the default
algorithm due to its simplicity. The current codes for IS algorithm are
reimplemented by Yuta Mori.
.B bwtsw
Algorithm implemented in BWT-SW. This method works with the whole human
genome, but it does not work with database smaller than 10MB and it is
usually slower than IS.

.B aln
bwa aln [-n maxDiff] [-o maxGapO] [-e maxGapE] [-d nDelTail] [-i
nIndelEnd] [-k maxSeedDiff] [-l seedLen] [-t nThrds] [-cRN] [-M misMsc]
[-O gapOsc] [-E gapEsc] [-q trimQual] <in.db.fasta> <in.query.fq> >

Find the SA coordinates of the input reads. Maximum
.I maxSeedDiff
differences are allowed in the first
.I seedLen
subsequence and maximum
.I maxDiff
differences are allowed in the whole sequence.

.TP 10
.B -n NUM
Maximum edit distance if the value is INT, or the fraction of missing
alignments given 2% uniform base error rate if FLOAT. In the latter
case, the maximum edit distance is automatically chosen for different
read lengths. [0.04]
.B -o INT
Maximum number of gap opens [1]
.B -e INT
Maximum number of gap extensions, -1 for k-difference mode (disallowing
long gaps) [-1]
.B -d INT
Disallow a long deletion within INT bp towards the 3'-end [16]
.B -i INT
Disallow an indel within INT bp towards the ends [5]
.B -l INT
Take the first INT subsequence as seed. If INT is larger than the query
sequence, seeding will be disabled. For long reads, this option is
typically ranged from 25 to 35 for `-k 2'. [inf]
.B -k INT
Maximum edit distance in the seed [2]
.B -t INT
Number of threads (multi-threading mode) [1]
Mismatch penalty. BWA will not search for suboptimal hits with a score
lower than (bestScore-misMsc). [3]
Gap open penalty [11]
Gap extension penalty [4]
Proceed with suboptimal alignments if there are no more than INT equally
best hits. This option only affects paired-end mapping. Increasing this
threshold helps to improve the pairing accuracy at the cost of speed,
especially for short reads (~32bp).
.B -c
Reverse query but not complement it, which is required for alignment in
the color space.
.B -N
Disable iterative search. All hits with no more than
.I maxDiff
differences will be found. This mode is much slower than the default.
.B -q INT
Parameter for read trimming. BWA trims a read down to
argmax_x{\\sum_{i=x+1}^l(INT-q_i)} if q_l<INT where l is the original
read length. [0]

.B samse
bwa samse [-n maxOcc] <in.db.fasta> <in.sai> <in.fq> > <out.sam>

Generate alignments in the SAM format given single-end reads. Repetitive
hits will be randomly chosen.
165 166 167 168 169

.TP 10
.B -n INT
170 171 172
Maximum number of alignments to output in the XA tag for reads paired
properly. If a read has more than INT hits, the XA tag will not be
written. [3]
173 174 175 176

.B sampe
177 178
bwa sampe [-a maxInsSize] [-o maxOcc] [-n maxHitPaired] [-N maxHitDis]
[-P] <in.db.fasta> <in1.sai> <in2.sai> <in1.fq> <in2.fq> > <out.sam>
179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194

Generate alignments in the SAM format given paired-end reads. Repetitive
read pairs will be placed randomly.

.TP 8
.B -a INT
Maximum insert size for a read pair to be considered being mapped
properly. Since 0.4.5, this option is only used when there are not
enough good alignment to infer the distribution of insert sizes. [500]
.B -o INT
Maximum occurrences of a read for pairing. A read with more occurrneces
will be treated as a single-end read. Reducing this parameter helps
faster pairing. [100000]
195 196 197 198 199
.B -P
Load the entire FM-index into memory to reduce disk operations
(base-space reads only). With this option, at least 1.25N bytes of
memory are required, where N is the length of the genome.
200 201 202 203 204 205 206 207 208 209
.B -n INT
Maximum number of alignments to output in the XA tag for reads paired
properly. If a read has more than INT hits, the XA tag will not be
written. [3]
Maximum number of alignments to output in the XA tag for disconcordant
read pairs (excluding singletons). If a read has more than INT hits, the
XA tag will not be written. [10]
210 211 212

213 214
.B bwasw
bwa bwasw [-a matchScore] [-b mmPen] [-q gapOpenPen] [-r gapExtPen] [-t
215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335
nThreads] [-w bandWidth] [-T thres] [-s hspIntv] [-z zBest] [-N
nHspRev] [-c thresCoef] <in.db.fasta> <in.fq>

Align query sequences in the <in.fq> file.

.TP 10
.B -a INT
Score of a match [1]
.B -b INT
Mismatch penalty [3]
.B -q INT
Gap open penalty [5]
.B -r INT
Gap extension penalty. The penalty for a contiguous gap of size k is
q+k*r. [2]
.B -t INT
Number of threads in the multi-threading mode [1]
.B -w INT
Band width in the banded alignment [33]
Minimum score threshold divided by a [37]
Coefficient for threshold adjustment according to query length. Given an
l-long query, the threshold for a hit to be retained is
a*max{T,c*log(l)}. [5.5]
.B -z INT
Z-best heuristics. Higher -z increases accuracy at the cost of speed. [1]
.B -s INT
Maximum SA interval size for initiating a seed. Higher -s increases
accuracy at the cost of speed. [3]
Minimum number of seeds supporting the resultant alignment to skip
reverse alignment. [5]

The output of the
.B `aln'
command is binary and designed for BWA use only. BWA outputs the final
alignment in the SAM (Sequence Alignment/Map) format. Each line consists

center box;
cb | cb | cb
n | l | l .
Col	Field	Description
1	QNAME	Query (pair) NAME
2	FLAG	bitwise FLAG
3	RNAME	Reference sequence NAME
4	POS	1-based leftmost POSition/coordinate of clipped sequence
5	MAPQ	MAPping Quality (Phred-scaled)
6	CIAGR	extended CIGAR string
7	MRNM	Mate Reference sequence NaMe (`=' if same as RNAME)
8	MPOS	1-based Mate POSistion
9	ISIZE	Inferred insert SIZE
10	SEQ	query SEQuence on the same strand as the reference
11	QUAL	query QUALity (ASCII-33 gives the Phred base quality)
12	OPT	variable OPTional fields in the format TAG:VTYPE:VALUE

Each bit in the FLAG field is defined as:

center box;
cb | cb | cb
c | l | l .
Chr	Flag	Description
p	0x0001	the read is paired in sequencing
P	0x0002	the read is mapped in a proper pair
u	0x0004	the query sequence itself is unmapped
U	0x0008	the mate is unmapped
r	0x0010	strand of the query (1 for reverse)
R	0x0020	strand of the mate
1	0x0040	the read is the first read in a pair
2	0x0080	the read is the second read in a pair
s	0x0100	the alignment is not primary
f	0x0200	QC failure
d	0x0400	optical or PCR duplicate

The Please check <> for the format
specification and the tools for post-processing the alignment.

BWA generates the following optional fields. Tags starting with `X' are
specific to BWA.

center box;
cb | cb
cB | l .
Tag	Meaning
NM	Edit distance
MD	Mismatching positions/bases
AS	Alignment score
X0	Number of best hits
X1	Number of suboptimal hits found by BWA
XN	Number of ambiguous bases in the referenece
XM	Number of mismatches in the alignment
XO	Number of gap opens
XG	Number of gap extentions
XT	Type: Unique/Repeat/N/Mate-sw
XA	Alternative hits; format: (chr,pos,CIGAR,NM;)*
337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377
XS	Suboptimal alignment score
XF	Support from forward/reverse alignment
XE	Number of supporting seeds

Note that XO and XG are generated by BWT search while the CIGAR string
by Smith-Waterman alignment. These two tags may be inconsistent with the
CIGAR string. This is not a bug.

.SS Alignment Accuracy
When seeding is disabled, BWA guarantees to find an alignment
containing maximum
.I maxDiff
differences including
.I maxGapO
gap opens which do not occur within
.I nIndelEnd
bp towards either end of the query. Longer gaps may be found if
.I maxGapE
is positive, but it is not guaranteed to find all hits. When seeding is
enabled, BWA further requires that the first
.I seedLen
subsequence contains no more than
.I maxSeedDiff
When gapped alignment is disabled, BWA is expected to generate the same
alignment as Eland, the Illumina alignment program. However, as BWA
change `N' in the database sequence to random nucleotides, hits to these
random sequences will also be counted. As a consequence, BWA may mark a
unique hit as a repeat, if the random sequences happen to be identical
to the sequences which should be unqiue in the database. This random
behaviour will be avoided in future releases.
By default, if the best hit is no so repetitive (controlled by -R), BWA
also finds all hits contains one more mismatch; otherwise, BWA finds all
equally best hits only. Base quality is NOT considered in evaluating
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398
hits. In paired-end alignment, BWA pairs all hits it found. It further
performs Smith-Waterman alignment for unmapped reads with mates mapped
to rescue mapped mates, and for high-quality anomalous pairs to fix
potential alignment errors.

.SS Estimating Insert Size Distribution
BWA estimates the insert size distribution per 256*1024 read pairs. It
first collects pairs of reads with both ends mapped with a single-end
quality 20 or higher and then calculates median (Q2), lower and higher
quartile (Q1 and Q3). It estimates the mean and the variance of the
insert size distribution from pairs whose insert sizes are within
interval [Q1-2(Q3-Q1), Q3+2(Q3-Q1)]. The maximum distance x for a pair
considered to be properly paired (SAM flag 0x2) is calculated by solving
equation Phi((x-mu)/sigma)=x/L*p0, where mu is the mean, sigma is the
standard error of the insert size distribution, L is the length of the
genome, p0 is prior of anomalous pair and Phi() is the standard
cumulative distribution function. For mapping Illumina short-insert
reads to the human genome, x is about 6-7 sigma away from the
mean. Quartiles, mean, variance and x will be printed to the standard
error output.
399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441

.SS Memory Requirement
With bwtsw algorithm, 2.5GB memory is required for indexing the complete
human genome sequences. For short reads, the
.B `aln'
command uses ~2.3GB memory and the
.B `sampe'
command uses ~3.5GB.

.SS Speed
Indexing the human genome sequences takes 3 hours with bwtsw
algorithm. Indexing smaller genomes with IS or divsufsort algorithms is
several times faster, but requires more memory.
Speed of alignment is largely determined by the error rate of the query
sequences (r). Firstly, BWA runs much faster for near perfect hits than
for hits with many differences, and it stops searching for a hit with
l+2 differences if a l-difference hit is found. This means BWA will be
very slow if r is high because in this case BWA has to visit hits with
many differences and looking for these hits is expensive. Secondly, the
alignment algorithm behind makes the speed sensitive to [k log(N)/m],
where k is the maximum allowed differences, N the size of database and m
the length of a query. In practice, we choose k w.r.t. r and therefore r
is the leading factor. I would not recommend to use BWA on data with
Pairing is slower for shorter reads. This is mainly because shorter
reads have more spurious hits and converting SA coordinates to
chromosomal coordinates are very costly.
In a practical experiment, BWA is able to map 2 million 32bp reads to a
bacterial genome in several minutes, map the same amount of reads to
human X chromosome in 8-15 minutes and to the human genome in 15-25
minutes. This result implies that the speed of BWA is insensitive to the
size of database and therefore BWA is more efficient when the database
is sufficiently large. On smaller genomes, hash based algorithms are
usually much faster.

.B `bwasw'
is designed for long-read alignment. The algorithm behind, BWA-SW, is
444 445 446
similar to BWT-SW, but does not guarantee to find all local hits due to
the heuristic acceleration. It tends to be faster and more accurate if
the resultant alignment is supported by more seeds, and therefore
BWA-SW usually performs better on long queries than on short ones.

On 350-1000bp reads, BWA-SW is several to tens of times faster than the
existing programs. Its accuracy is comparable to SSAHA2, more accurate
than BLAT. Like BLAT, BWA-SW also finds chimera which may pose a
challenge to SSAHA2. On 10-100kbp queries where chimera detection is
important, BWA-SW is over 10X faster than BLAT while being more
454 455

BWA-SW can also be used to align ~100bp reads, but it is slower than
457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483
the short-read algorithm. Its sensitivity and accuracy is lower than
SSAHA2 especially when the sequencing error rate is above 2%. This is
the trade-off of the 30X speed up in comparison to SSAHA2's -454 mode.

BWA website <>, Samtools website

Heng Li at the Sanger Institute wrote the key source codes and
integrated the following codes for BWT construction: bwtsw
<>, implemented by Chi-Kwong Wong at
the University of Hong Kong and IS
<> originally proposed by Nong Ge
<> at the Sun Yat-Sen University and
implemented by Yuta Mori.

The full BWA package is distributed under GPLv3 as it uses source codes
from BWT-SW which is covered by GPL. Sorting, hash table, BWT and IS
libraries are distributed under the MIT license.
If you use the short-read alignment component, please cite the following
Li H. and Durbin R. (2009) Fast and accurate short read alignment with
484 485 486 487 488 489
Burrows-Wheeler transform. Bioinformatics, 25, 1754-60. [PMID: 19451168]
If you use the long-read component (BWA-SW), please cite:
Li H. and Durbin R. (2010) Fast and accurate long-read alignment with
Burrows-Wheeler transform. Bioinformatics. [PMID: 20080505]
490 491 492

BWA is largely influenced by BWT-SW. It uses source codes from BWT-SW
and mimics its binary file formats; BWA-SW resembles BWT-SW in several
494 495 496
ways. The initial idea about BWT-based alignment also came from the
group who developed BWT-SW. At the same time, BWA is different enough
from BWT-SW. The short-read alignment algorithm bears no similarity to
Smith-Waterman algorithm any more. While BWA-SW learns from BWT-SW, it
498 499 500 501 502 503 504 505 506 507 508 509 510 511 512
introduces heuristics that can hardly be applied to the original
algorithm. In all, BWA does not guarantee to find all local hits as what
BWT-SW is designed to do, but it is much faster than BWT-SW on both
short and long query sequences.

I started to write the first piece of codes on 24 May 2008 and got the
initial stable version on 02 June 2008. During this period, I was
acquainted that Professor Tak-Wah Lam, the first author of BWT-SW paper,
was collaborating with Beijing Genomics Institute on SOAP2, the successor
to SOAP (Short Oligonucleotide Analysis Package). SOAP2 has come out in
November 2008. According to the SourceForge download page, the third
BWT-based short read aligner, bowtie, was first released in August
2008. At the time of writing this manual, at least three more BWT-based
short-read aligners are being implemented.

The BWA-SW algorithm is a new component of BWA. It was conceived in
November 2008 and implemented ten months later.