If you use CAP3, a citation of this paper would be appreciated:
Huang, X. and Madan, A. (1999)
CAP3: A DNA Sequence Assembly Program,
Genome Research, 9: 868-877.
This page was adapted from the doc file provided with the source distribution.
We have made the following improvements to the CAP sequence
- Use of forward-reverse constraints to correct assembly errors and link contigs.
- Use of base quality values in alignment of sequence reads.
- Automatic clipping of 5' and 3' poor regions of reads.
- Generation of assembly results in ace file format for Consed.
- CAP3 can be used in GAP4 of the Staden package.
The improved program is named CAP3. These improvements allow CAP3 to take longer sequences of higher errors and produce more accurate consensus sequences.
Use of constraints in layout generation
A forward-reverse constraint is often produced by sequencing both ends of a subclone. A forward-reverse constraint specifies that the two reads should be on the opposite strands of the DNA molecule within a specified range of distance.
CAP3 makes use of a large number of forward-reverse constraints to locate and correct errors in layout of sequence reads. This capability allows CAP3 to address assembly errors due to repeats.
CAP3 also uses constraints to link contigs separated by a gap. This feature provides useful information to sequence finishers.
The algorithm used in CAP3 is designed to tolerate wrong constraints, which are due to errors in naming and lane tracking.
Use of quality values in alignment
CAP3 makes use of base quality values in constructing an alignment of sequence reads and generating a consensus sequence for each contig.
This allows the program to use both base quality values and the depth of coverage at a position to improve the accuracy in generating a consensus base at the position.
The alignment method in CAP3 is very tolerable of reads of high sequencing errors.
Automatic clipping of 5' and 3' poor regions
CAP3 clips 5' and 3' poor regions of reads and uses only good regions of reads in assembly. Thus there is no need to perform clipping in advance. Note that vector sequences in reads must be masked before using CAP3.
Input to CAP3
CAP3 takes as input a file of sequence reads in FASTA format. If the names of reads contain a dot ('.'), CAP3 requres that the names of reads sequenced from the same subclone contain the same substring up to the first dot. CAP3 takes two optional files: a file of quality values in FASTA format and a file of forward-reverse constraints.
The file of quality values must be named "xyz.qual", and the file of forward-reverse constraints must be named "xyz.con", where "xyz" is the name of the sequence file. CAP3 uses the same format of a quality file as Phrap.
Each line of the constraint file specifies one
forward-reverse constraint of the form:
ReadA ReadB MinDistance MaxDistance
where ReadA and ReadB are names of two reads, and MinDistance and MaxDistance are distances (integers) in base pairs. The constraint is satisfied if ReadA in forward orientation occurs in a contig before ReadB in reverse orientation, or ReadB in forward orientation occurs in a contig before ReadA in reverse orientation, and their distance is between MinDistance and MaxDistance. CAP3 works better if a lot more constraints are used.
We have a separate program named "formcon" to generate a constraint file from the sequence file. The program takes an input file of fragments in FASTA format and two integers (minimum distance and maximum distance in bp). The minimum distance and maximum distances specify a lower and a upper limit on the subclone length, respectively. It produces a file of forward-reverse constraints for CAP3. It is assumed that a pair of forward and reverse reads must contain a dot in their names and a pair of forward and reverse reads have a common name up to the first dot. Because CAP3 uses reads whose ends are clipped, instead of raw reads, to measure their distance, the distance seen by CAP3 could be different from the insert size by 1000 to 1500 bp. For example, if the insert size is 2000 to 3000 bp, we recommend that you use 500 for the minimum distance and 4000 for the maximum distance. The results are in the file with name ending in ".con".
Output from CAP3
Assembly results in CAP format go to the standard output and need to be directed to a file. Note that clipped 5' and 3' sequences of reads are not shown in CAP3 format output.
CAP3 also produces assembly results in ace file format (".ace"). This allows CAP3 output to be viewed in Consed. Note that clipped 5' and 3' sequences of reads are shown in ace format output.
CAP3 saves consensus sequences in file ".contigs" and their quality values in file ".contigs.qual". Reads that are not used in assembly are put in file ".singlets". Additional information about assembly is given in file ".info".
The CAP3 program reports whether each constraint is
satisfied or not. The report is in file ".results". A sample
report file is given here (spacing will be different):
CPBKY55.F CPBKY55.R 500 6000 3210 satisfied
The first four columns are simply taken from the constraint file.
CPBKY92.F CPBKY92.R 500 6000 497 unsatisfied in distance
CPBKY28.F CPBKY28.R 500 6000 unsatisfied
CPBKY56.F CPBKY56.R 500 6000 10th link between CPBKI23.F+ and CPBKT37.R-
CPBKY70.F CPBKY70.R 500 6000 4th overlap between CPBKM47.F+ and CPBKN28.R-
- indicates that the constraint is satisfied, where the actual distance between the two reads is given on the fifth column.
- indicates that the constraint is not satisfied in distance, that is, the two reads in opposite orientation occur in the same contig, but their distance (given on the fifth column) is out of the given range.
- indicates that the constraint is not satisfied.
- indicates that this constraint is a 10th one that links two contigs, where the 3' read of one contig is "CPBKI23.F" in plus orientation and the 5' read of the other is "CPBKT37.R" in minus orientation. The information suggests that the two contigs should go together in the gap closure phase.
- indicates that the constraint is a 4th constraint supporting an overlap between CPBKM47.F and CPBKN28.R. The overlap is not implemented in the current the assembly.
CAP3 takes 20 to 60 minutes to assemble a cosmid or BAC data set on a Sun Ultra1 workstation.
cap3 File_of_reads [options]
where "File_of_reads" is a file of DNA reads in FASTA format. If "File_of_reads" is named 'xyz', then the file of quality values must be named 'xyz.qual', and the file of constraints named 'xyz.con'.
Options (default values)
-a N ..specify band expansion size N > 10 (20)
-b N ..specify base quality cutoff for differences N > 15 (20)
-c N ..specify base quality cutoff for clipping N > 5 (12)
-d N ..specify max qscore sum at differences N > 100 (200)
-e N ..specify extra number of differences N > 10 (20)
-f N ..specify max gap length in any overlap N > 10 (300)
-g N ..specify gap penalty factor N > 0 (6)
-h N ..specify max overhang percent length N > 5 (20)
-i N ..specify segment pair score cutoff N > 20 (40)
-j N ..specify chain score cutoff N > 30 (80)
-k N ..specify end clipping flag N >= 0 (1)
-m N ..specify match score factor N > 0 (2)
-n N ..specify mismatch score factor N < 0 (-5)
-o N ..specify overlap length cutoff > 15 (40)
-p N ..specify overlap percent identity cutoff N > 65 (90)
-r N ..specify reverse orientation value N >= 0 (1)
-s N ..specify overlap similarity score cutoff N > 250 (900)
-t N ..specify max number of word occurrences N > 30 (500)
-u N ..specify min number of constraints for correction N > 0 (4)
-v N ..specify min number of constraints for linking N > 0 (2)
-w N ..specify file name for clipping information (none)
-x N ..specify prefix string for output file names (cap)
-y N ..specify clipping range N > 5 (100)
-z N ..specify min no. of good reads at clip pos N > 0 (2)
If no quality file is given, then a default quality value of 10 is used for each base.
The following sections explain the parameters of CAP3.
Clipping of poor regions
If the option
-k 0 is given, then no read
end is clipped and the whole read is used in
assembly. Otherwise, the following procedure is used to
determine and clip poor read ends.
CAP3 computes clipping positions of each read using both base quality values and similarity information. Clipping of a poor end region of a read f is controlled by three parameters: quality value cutoff qualcut, clipping range crange, and depth of good coverage gdepth. The value for qualcut can specified with the "-c" option, the value for crange with the "-y" option, and the value for gdepth with the "-z" option.
If there are quality values, CAP3 computes two positions qualpos5 and qualpos3 of read f such that the region of read f from position qualpos5 to position qualpos3 consists mostly of quality values greater than qualcut. If there are no quality values, then qualpos5 is set to 1 and qualpos3 is set the length of read f.
The range for the left clipping position of read f is from 1 to qualpos5 + crange.
The range for the right clipping position of read f is from qualpos3 - crange to the end of read f. The minimum depth of good coverage at the left and right clipping positions of read f is expected to be gdepth.
Let realdepth5 be the maximum real depth of coverage for the initial region of read f ending at position qualpos5 + crange. Let depth5 be the smaller of realdepth5 and gdepth. If depth5 is 0, then left clipping position of read f is set to qualpos5 by CAP3. The given value for the parameter crange may be too small for read f. CAP3 reports at the start of a .info file that "No overlap is found in the given 5' clipping range for read f." If there are overlaps beyond the given 5' clipping range for read f, CAP3 reports a new clipping range for each overlap. One of the reported range values can be used as a new value for the parameter crange for read f. If CAP3 reports "No overlap is found ... for read f", then read f is not used in assembly. A larger clipping range has to be given to use read f in assembly.
If depth5 is greater than 0, the left clipping position of read f is the smallest position x such that x is less than qualpos5 + crange and the region of read f beginning at position x is similar to depth5 other reads. The right clipping position of read f is computed similarly by CAP3. Larger values for the parameters crange and gdepth result in more aggressive clipping of poor end regions. A larger value for crange allows CAP3 to search for the left clipping position in a larger area. A larger value for gdepth may cause CAP3 to clip more bases so that the resulting good portion of read f is similar to more reads.
The user may provide specific values for the parameters
crange and gdepth for individual reads in a file. Each
line in the file has the following format:
ReadName crange5 gdepth5 crange3 gdepth3
where ReadName is the name of a read, crange5 & gdepth5 are values for the 5' end, and crange3 & gdepth3 are for the 3' end. The file is given to CAP3 with the "-w" option.
Band of diagonals
The program determines a minimum band of diagonals for an overlapping alignment between two sequence reads. The band is expanded by a number of bases specified by the user with option "-a".
Quality difference score of an overlap
Overlaps between reads are evaluated by many measures. The first measure is based on base quality. If an overlap contains lots of differences at bases of high quality, then the overlap is removed. Specifically, let b be the base quality cutoff value and let d be the maximum difference score. The values for the two parameters can be set using the "-b" and "-d" options. If the overlap contains a difference at bases of quality values q1 and q2, then the score at the difference is max(0, min(q1, q2) - b). The difference score of an overlap is the sum of scores at each difference. For example, an overlap contains two differences, one at bases of quality values 15 and 30 and the other at bases of quality values 40 and 50. With b = 20, the difference score of the overlap is 0 + 20 = 20. If the difference score of an overlap exceeds d, then the overlap is removed. With b = 20, an overlap with 15 differences at bases of quality values 40 or higher has a difference score of at least 300 and is removed if d = 250.
Number of differences in an overlap
The second measure looks at the number of differences in an overlap. If the number of differences in an overlap is higher than expected, than the overlap is removed. Let an integer e be the maximum number of extra differences. Consider an overlap between reads f and g. Let d1 be the estimated number of sequencing errors for the region of f involved in the overlap and let r2 be that for the region of g involved in the overlap. If the observed number of differences in the overlap is greater than r1 + r2 + e, then the overlap is removed. The value for the parameter e can be changed using the "-e" option. The expected number of differences in the overlap is about r1 + r2. Giving a smaller value to e causes more overlaps to be removed.
Similarity score of an overlap
The third measure is based on overlap similarity score. The similarity score of an overlapping alignment is defined using base quality values. Let m be the match score factor, let n be the mismatch score factor, and let g be the gap penalty factor. Values for these parameters can be set with the "-m", "-n", and "-g" options. A match at bases of quality values q1 and q2 is given a score of m * min(q1,q2). A mismatch at bases of quality values q1 and q2 is given a score of n * min(q1,q2). A base of quality value q1 in a gap is given a score of -g * min(q1,q2), where q2 is the quality value of the base in the other sequence right before the gap. The score of a gap is the sum of scores of each base in the gap minus a gap open penalty. The similarity score of an overlapping alignment is the sum of scores of each match, each mismatch, and each gap. With m = 2, an overlap that consists of 25 matches at bases of quality value 10 has a score of 500. If the similarity score of an overlap is less than the overlap similarity score cutoff s, then the overlap is removed.
Length and percent identity of an overlap
The fourth requirement for an overlap is that the length in bp of the overlap is no less than the value of the minimum overlap length cutoff parameter. The value for this parameter can be changed with the "-o" option. The fifth requirement for an overlap is that the percent identity of the overlap is no less than the minimum percent identity cutoff. The value for this parameter can be changed with the "-p" option. A value of 75 for p means 0.75 or 75%.
Maximum length of gaps in an overlap
The program provides a parameter (-f option) for the user to reject overlaps with a long gap. Let an integer f be the maximum length of gaps allowed in any overlap. Then any overlap with a gap longer than f is rejected by the program. The value for this parameter can be changed using the "-f" option. Note that a small value for this parameter may cause the program to remove true overlaps and to produce incorrect results. The "-f" option may be used by the user to split reads from alternative splicing forms into separate contigs. Geo Pertea at TIGR suggested that this option be added to the program.
Overhang percent length of an overlap
The total length of the different overhang regions in an overlap is controlled with the -h option. An overhang region in an overlap is a different terminal region before or after the overlap. The overhang percent length of an overlap is 100 times the total length of the different overhang regions in the overlap divided by the length of the overlap. Overlaps with an overhang percent length greater than the maximum cutoff are rejected.
The default values for some of the parameters are
selected for assembly of regular reads of lengths 500 to
1000 bp. For assembly of short reads of lengths 20 bp,
the following options should be used to change the
values for those parameters accordingly.
-i 30 -j 31 -o 18 -s 300
Note that using short reads increases the likelihood of producing assemblies with false joins. Below we explain the options for short reads. Overlaps between reads are quickly computed by finding segment pairs (ungapped alignments) and combining segment pairs into chains. The -i option is used to specify a score cutoff on segment pairs. The score of a segment pair with 19 base matches and 1 base mismatch is 2 * 19 + (-5) * 1 = 33, where each base match is given a score of 2 and each mismatch is given a score of -5. The -j option is used to specify a score cutoff on chains of segment pairs, where the score of a chain is the sum of scores of each segment pair minus penalties for gaps between segment pairs. The score of a chain consisting of one segment pair is simply the score of the segment pair.
After a high scoring chain of segment pairs between two reads is computed, an overlap between the reads is computed as an optimal local alignment between the reads, where the chain is used to limit the computation to a small area of the dynamic programming matrix. Unlike the scores of segment pairs and chains, the score of an overlap is weighted by base quality values. Thus, an overlap with 19 base matches, 1 base mismatch, and 0 gap has a score of 10 * [2 * 19 + (-5) * 1] = 330, assuming that each base has a quality value of 10. The -o option is used to specify a length cutoff on overlaps, whereas the -s option is used to specify a score cutoff on overlaps.
Assembly of reads in forward orientation only
The "-r" option is used to let CAP3 know whether to consider reads in reverse orientation for assembly. The default value for the option is 1, meaning that reads in reverse orientation are also considered for assembly. Specifying zero as "-r 0" instructs CAP3 to perform assembly of reads in forward orientation only. This option was suggested by Patrick Schnable's lab.
Max number of word matches
This parameter (option -t) allows the user to trade off the efficiency of the program for its accuracy. For a read f, the program computes overlaps between read f and other reads by considering short word matches between read f and other reads. A word match is examined to see if it can be extended into a long overlap. If read f has overlaps with many other reads, then read f has many short word matches with many other reads. This parameter gives an upper limit, for any word, on the number of word matches between read f and other reads that are considered by the program. Using a large value for this parameter allows the program to consider more word matches between read f and other reads, which can find more overlaps for read f, but slows down the program. Using a small vlaue for this parameter has the opposite effect. A large value may be used if the depth of coverage is high for the data set. For example, a value of 150 is used for a data set with a maximum depth of coverage of 30, and a value of 500 for a data set with a maximum depth of coverage of 100. Using a very large value may cause the program to run forever or run out of memory.
Corrections to an assembly are made using forward-reverse constraints. Let an integer u be the minimum number of constraints for correction. Consider an alternative overlap between two reads f and g. Assume that f is in contig C1 and that g is in contig C2. If the number of unsatisfied constraints that support the overlap between f and g is greater than the value of the u parameter plus the number of satisfied constraints that support the current joins involving f and g, then the current joins involving f and g are disconnected and the overlap between f and g is implemented. The value for this parameter can be changed with the "-u" option.
Contigs that are linked by forward-reverse constraints are reported. The minimum number of constraints for reporting a link between two contigs is specified with the "-v" option.
Output file names
The names of all output files contain a substring "cap". A different substring can be specified with the "-x" option. This feature was suggested by Harley Gorrell.
I thank John Quackenbush, Geo Pertea, and Feng Liang for many suggestions to improve CAP3, Jun Qian for producing output in ace format and other help, Kathryn Beal for incorporating CAP3 in GAP4, Tim Hunkapiller and Granger Sutton for discussion, Bruce Roe and Granger Sutton for providing sequence data sets, Sanzhen Liu and Pat Schnable for suggesting the options -i, -j, -k. This project was supported by NIH Grant R01HG01502-02 from NHGRI.
Users may direct questions to firstname.lastname@example.org.