This is a modified version of VSEARCH. See https://github.com/torognes/vsearch to know about VSEARCH.
Firstly, the --fastq_join2 option was added. The --fastq_join option can be concatenated paired-end reads as shown below.
forward_read -> - padding - <- reverse_read
However, the --fastq_join2 option can be concatenated paired-end reads like the following.
<- reverse_read - padding - forward_read ->
Secondly, the --idoffset option was added. If you specify --idoffset N (N must be integer), the identity will be calculated based on the following definition.
100 * (the number of matches - N) / (alignment length - N)
This argument should be useful for clustering of non-overlapped paired-end reads as explained below.
Assuming you constructed concatenated sequences by the --fastq_join2 option as shown in above, the clustered results will be OVERclustered without the --idoffset option because of padding sequence. In such cases, giving the length of padding sequence as idoffset will produce correctly clustered results. After the clustering, the representative or consensus sequences can be divided into forward and reverse parts using padding sequence.