Maybe you can try to merge the read1 (forward) and read2 (reverse) sequences using multiple_join_paired_ends.py:

multiple_join_paired_ends.py -i input_directory -o output_directory --read1_indicator _R1 --read2_indicator _R2

You can also check the options on the web (http://qiime.org/scripts/multiple_join_paired_ends.html), but I think you don't need to convert to the reverse complementary.

Good luck!.

Hi, it was paired end sequencing you have data for? Then you should proceed with joining the forward and reverse reads.

Thanks José and Lisa. I am Juhee's colleague.

Our sequencing data is NOT paired. The forward and reverse reads are unparallel - each orientation makes up about 50% of the results, but not exactly. This is because we PCR first, and then enzymatically attach barcodes and adaptors at random orientation.

Any further comments/recommendations would be grateful!

I'm about to try the same thing. I have forward and reverse reads mixed in a single .fastq file from Ion Torrent PGM.

So far, I could separate the forward and reverse reads from the mixed .fastq files using Split_on_Primer (https://github.com/Y-Lammers/Split_on_Primer)

Now I'm looking for a tool that allows me to merge the reads that are not in parallel and that does not give me any label issues. In fact, neither the -fastq_mergepairs in uparse9.2 nor the make.contigs of mothur works. 

Would be great if anyone could give a hint on a tool that does it

@Seung Gu Shin & Juhee Shin: It depends on the HV region you targeted and the read length. If the PCR fragment is short enough so that you will have most probably sequenced the whole fragment for the majority of your reads, then you will have in essence two types of reads (lets assume you have targeted V1-V2 and used 8F and 361R primer):

5' 8F V1V2 361R 3'

5' 361R_revcomp V2V1 8F_revcomp 3'

Then you only need to search for the reverse primer in reverse complement direction at the beginning in your reads (you would need to search and trim primers anyway) using e.g. cutadapt (using anchor flags and some dissimilarities allowed) and, if identified, reverse complement the whole read. Then you have all reads in same orientation. If you want to be very sure, you also search for the revcomp forward primer at the 3'-end. Of course, you would need to search and trim for the other primers of the remaining sequences.

It may be wise to verify how good this carries out on shorter sequences, if e.g. the 3'-end primer cannot be identified because it was not sequenced. Reverse complementing such a read would result in a pool of reads starting at different position. This would make any downstream processing difficult (e.g. OTU clustering could be inflated). Maybe it would make sense to include only those reads, where both (the forward and the reverse) primer can be identified (in parts, at least) - if this is not removing too many reads from the dataset, of course.

I hope this helps. This is based on experience, but this experience is old. I haven't used PGM amplicons for  years now, MiSeq data is much more easy to handle.

- Sebastian

@Sebastian

Many thanks for your suggestion. I am working with exactly the region you described and unfortunately, most reads are not full length. Especially after quality filtering the reads end up being way shorter. Therefore, it would make sens, if possible, to merge the Forward and Reverse reads.

If not python / perl or other script is available that does it in unpaired reads, the only alternative I have spotted that allows me not to lose the reverse reads, is to treat the forward and reverse reads separately and finally merge the otu table after mapping.

@Marco: I'm not sure whether we are talking about the same thing or not. The problem described by Seung Gu Shin & Juhee Shin does not relate to problems merging the forward and reverse read of a paired-end sequencing.

So, merging is only possible if PE sequencing was done (as opposed to single-end sequencing). And for that it doesn't matter if the reads are all the same orientation (because the individual read pairs will be oriented equally, the only thign that matters for this step). You just merge them (the forward and the reverse mate of this pair) and then you treat them like described above, i.e. search for those ones who are oriented 3'-5' (via the PCR primers in opposite orientation) and reverse complement them. Done. If both reads are not long enough to merge them (because there is no overlapping region), then there is not much you can do about this.

If you don't have PE reads, then you can't, of course, just stitch ambiguously reads together, neither if half of them was sequenced on the opposing stand.

Anyhow, I would perform the orientation 'correction' and primer removal at the beginning and not the end of quality filtering, because trimming reads on quality may make it very problematic to identify 3'-end primers., and so on.

@Sebastian: thanks for the quick response. I see your point. It makes sense to reverse complement the reverse sequences and process them together with the forward after primer trimming.

However, unfortunately most reverse sequences do not reach V1 or only partly cover V1, which is why there is a significant difference in sequence variability between the forward as compared to the reverse reads. The forward reads cover V1 and V2 and mostly are truncated in the conserved region, while the reverse is truncated either before or in V1.

I am not sure how this could affect OTU-clustering and downstream analyses but need to check. Anyway, many thanks for your suggestions.