We will be offering an R workshop December 18-20, 2019. Learn more.
The align.seqs command aligns a user-supplied fasta-formatted candidate sequence file to a user-supplied fasta-formatted template alignment. The general approach is to i) find the closest template for each candidate using kmer searching, blastn, or suffix tree searching; ii) to make a pairwise alignment between the candidate and de-gapped template sequences using the Needleman-Wunsch, Gotoh, or blastn algorithms; and iii) to re-insert gaps to the candidate and template pairwise alignments using the NAST algorithm so that the candidate sequence alignment is compatible with the original template alignment. We provide several alignment databases for 16S and 18S rRNA gene sequences that are compatible with the greengenes or SILVA alignments; however, custom alignments for any DNA sequence can be used as a template an users are encouraged to share their alignment for others to use. In general the alignment is very fast - we are able to align over 186,000 full-length sequences to the SILVA alignment in less than 3 hrs with a quality as good as the SINA aligner. Furthermore, this rate can be accelerated using multiple processors. While the aligner doesn't explicitly take into account the secondary structure of the 16S rRNA gene, if the template database is based on the secondary structure, then the resulting alignment will at least be implicitly based on the secondary structure. To demonstrate the various features of this command, we will use the AbRecovery dataset.
The align.seqs command requires the user to enter a candidate and template file. First, obtain an alignment database and make sure that it is located in the same folder as your candidate file and where you are running mothur from. We will use the greengenes alignment with 7,682 positions. To run the aligner enter the following command:
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta)
This will generate the following output:
Reading in the core_set_aligned.imputed.fasta template sequences... DONE. Generating the core_set_aligned.imputed.8mer database... DONE.
It took 7 secs to align 242 sequences
Note that once the database was generated, it took 7 seconds. The command will generate two files - abrecovery.align and abrecovery.align.report. The align file is a fasta-formatted file that contains the 242 aligned sequences. We put periods (i.e. '.') leading up to the first base in the sequence and following the last base of the sequence. The report file contains information about the quality of the alignment. For example...
QueryName QueryLength TemplateName TemplateLength SearchMethod SearchScore AlignmentMethod QueryStart QueryEnd TemplateStart TemplateEnd PairwiseAlignmentLength GapsInQuery GapsInTemplate LongestInsert SimBtwnQuery&Template AY457915 501 82283 1525 kmer 89.07 needleman 5 501 1 499 499 2 0 0 97.60
In this example we see that candidate sequence AY457915 was 501 bases long and we used kmer searching (i.e. using 8mers) to find template sequence 82283 as the best match. Sequence 82283 had 89% of the 8mers found in AY457915. Next, we see that we used the needleman alignment method, which resulted in a pairwise alignment that was 501 characters long. In this example we see that the alignment actually starts at candidate sequence position 5. This occurred because the sequence actually has vector sequence at the 5' end, which is not represented in the reference alignment, which includes the traditional 27f and 1492r primer sites. Next, we see that during the pairwise alignment stage, 2 gaps were entered into the candidate sequence and none were introduced into the template sequence - none of the template gaps needed to be corrected with the NAST algorithm. Finally, the aligned candidate sequence was 97.6% identical to the template sequence (including gaps).
By default, align.seqs will use kmer searching with 8mers and will use the Needleman-Wunsch pairwise alignment method with a reward of +1 for a match and penalties of -1 and -2 for a mismatch and gap, respectively. Also, only one processor will be used with the default settings.
mothur offers three methods of finding the template sequence - kmer searching, blast, and suffix tree searching. Our experience has shown that kmer searching is the fastest and best method of searching for a template sequence. The default is to use kmers:
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta, search=kmer)
Alternatively, you can use blastn:
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta, search=blast)
If you use blast, you need to install the blast folder in your source code folder or with the executable so that the path from the mothur executable to the blast executables is /blast/bin/. For the search, you need to have formatdb and blastall in the bin folder. Also, be forewarned that to make this as fast as possible, we chose a word size of 28 (-W 28), which has the unintended consequence of returning more "no significant" searches.
The final option is to use suffix trees for searching:
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta, search=suffix)
This approach is fairly comparable to blast in terms of quality and speed.
If you use the kmer option, you can change the size of kmers that are used in align.seqs. By default this is set to use 8mers. Alternatively, you can use kmers ranging in size between 5 and 12. We have noticed that the best kmer size is typically the one that is the fastest. For full length alignment databases, kmers of 8 or 9 are generally appropriate. If you use another database, be sure to experiment with the settings to see what works best. To change the kmer size:
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta, ksize=6)
The align.seqs command allows you to select between three alignment methods - blastn, gotoh, and needleman - needleman is the default setting:
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta, align=needleman)
The needleman algorithm penalizes the same amount for opening and extending a gap. Alternatively, you could use the gotoh algorithm, which charges a different penalty for opening (default=-2) and extending (default=-1) gaps:
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta, align=gotoh)
Our experience has shown that the added parameters in the gotoh algorithm do not improve the pairwise alignment and increases the time required for the alignment. Finally, blastn can be used as a heuristic approach to the gotoh alignment:
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta, align=blast)
As with the search option, if you use blast, you need to install the blast folder in your source code folder or with the executable so that the path from the mothur executable to the blast executables is /blast/bin/. For the align=blast option, you need to have bl2seq in the bin folder. In our implementation, blast is the slowest option of the three and also generates the worst alignments. The quality suffers particularly because it generates a local alignment (needleman and gotoh are global) and will truncate the alignment if the similarity drops below a threshold.
match, mismatch, gapopen, and gapextend
In the pairwise alignment portion of the aligning procedure, the default reward for a match is +1 and the penalties for a mismatch, opening and extending a gap are -1, -2, and -1. Our experience has shown that these produce the best alignments for 16S rRNA gene sequences. You are encouraged to play around with these to suit your own purposes as shown below:
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta, align=gotoh, match=1, mismatch=-3)
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta, align=gotoh, gapopen=-5)
Keep in mind that if you are using the align=blast option, blast will limit the combinations of match, mismatch, gapopen, and gapextend that you can use. Hopefully, we've scared you off of using blast at all so that this won't be an issue.
flip and threshold
The threshold parameter is used to specify a cutoff at which an alignment is deemed 'bad' and the reverse complement may be tried. The default threshold is 0.50, meaning if 50% of the bases are removed in the alignment process. The flip parameter is used to specify whether or not you want mothur to try the reverse complement of a sequence if the sequence falls below the threshold. The default is false. If the flip parameter is set to true the reverse complement of the sequence is aligned and the better alignment is reported.
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta, flip=t)
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta, flip=t, threshold=0.75)
If you have a Windows computer, move on, this feature doesn't apply to you, unless you are using the mpi-enabled version. If you are using the mpi-enabled version, processors is set to the number of processes you have running. If you're one of the cool kids, you get to use the processors option, which enables you to accelerate the alignment by using multiple processors. You are able to use as many processors as your computer has with the following option:
mothur > align.seqs(candidate=abrecovery.fasta, template=core_set_aligned.imputed.fasta, processors=2)
Running this command on my laptop doesn't exactly cut the time in half, but it's pretty close. There is no software limit on the number of processors that you can use.
Differences in implementation
The most similar implementation of the align.seqs command is the online greengenes aligner. The NAST paper describes using 7mer searching and blast alignments, which could align 10 sequences per minute. In contrast, align.seqs can align about 18 16S rRNA gene sequences per second to a 50,000 character alignment and 22 sequences per second to a 8,000 character alignment. Since the publication of the NAST aligner, the strategy has changed to using blast to find a template and the alignment. To select the template, the greengenes aligner selects the longest among the top-ten most similar template sequences.
Another important difference between align.seqs and the other aligners that are available is that the aligner is not tied to any one alignment. You could use the greengenes, SILVA, or RDP alignments. You could even design a recA alignment and use that.
There are several considerations to remember when aligning your sequences:
- The number of sequences to be aligned. If you use a program such as muscle or clustal, doubling the number of sequences will increase the amount of time required by 16-fold and the amount of RAM required by 4-fold. In contrast, doubling the number of sequences will double the amount of time required and require no extra RAM.
- The number of sequences in the template database. Since the templates are stored in RAM, you need to consider how many sequences to use. Our version of the SILVA SEED database can be loaded in less than 2GB of RAM. The number of sequences in the database will have an effect on search times. Roughly speaking, doubling the number of template sequences should double the amount of time required to find the best match.
- The length of candidate sequences. Doubling the sequence length will approximately double the time required to find the best match. In the alignment step, doubling the length of the sequences will quadruple the amount of time required. For comparison, align.seqs will align 18 full-length sequences per second with 8mers and the needleman algorithm and 31 750-bp fragments (27f-787r) per second. There are optimizations that we do in the kmer searching which probably explain the differences.
- The length of the alignment. Using a 50,000 character alignment vs. a 7,700 character alignment means that an extra 42,300 gaps need to be inserted into the alignment, which is pretty easy, but still takes time. The bigger problem is the effect that it has on storing the database in RAM and for downstream processing. Using the filter.seqs command we are able to remove the vertical gaps from the template database and the aligned candidate sequences. When we aligned the 186,000 full-length sequences to a version of the SILVA database with all vertical gaps removed, the sequence quality was just as high as with the full alignment. We were able to align 22 sequences per second with the filtered template database vs. 18 sequences per second with the full database.
For purposes of comparison, when we aligned 186,000 full-length sequences to the SILVA template library the search with 8mers required 3,800 seconds and the alignment required about 6,500 seconds.
Sub-regions of the 16S rRNA gene align very well, when the template database is well curated in that region. Depending on the region you use it may be necessary to tweak the search and alignment parameters. The align.seqs defaults work well for aligning variable region sequences to a full-length alignment template database. If you are analyzing V6 pyrotags, you might consider using ksize=9. We have experimented with alignments using region-specific template databases and have found that the alignment quality is not significantly better; however, there is a considerable speed up. For example using 9mer searching with needleman to align 186,000 V6 fragments against the full-length database allowed us to align 78 sequences per second. In contrast, aligning the same tags against a template alignment of just the V6 region with 5mers allowed us to align 145 sequences per second. There was no significant difference in alignment quality.
There are two very important things to consider:
- Your alignment is only as good as the database you are aligning to. The old adage - garbage in, garbage out - hold true in sequence alignment. If you are interested in a particular variable region, then it would be worth your while to make sure that the database you are using is well aligned in that region. This is the only reason we might suggest the SILVA alignment over the greengenes alignment.
- Don't instinctively trust the alignment you get from any aligner. It would be worth your while to import your sequences into an alignment editor that takes into account the secondary structure (e.g. ARB) to make sure that things look right.