Bowtie2 End-to-End or Local alignment mode, Which one i should choose for downstream bacteria SNPs and Indels calling

Question

[Calvin]

I have used bowtie2 for both end-to-end and local alignment (sensitive mode) using raw data (Preprocessed) from illumina Miseq. the read length are in the range of 36 to 150. The results shows that Local alignment gives around 92% alignment rates but end-to-end alignment gives around 64%. Which one i should choose for bacteria SNPs and Indels calling?

Answer 1

[Devon Ryan]

Use local alignment, since it gives you 50% more alignments to work with.

[JV]

Could be right (especially for unprocessed seqeuncing data), but just as a thought (because I'm not sure myself): wouldn't the end-to-end alignments be more trustworthy (since they are based only on reads that map in their entire length)? I mean, couldn't local alignments result in some fake alignment-hits (based on partial similarities that do not extend to the full read length) that could complicate the SNP calling? (of course only assuming quality trimmed data. Since the OP calls his data "raw" and "preprocessed" at the same time, I'm not quite sure what applies here)

[Devon Ryan]

They'll both result in false alignments. You'll get more artificial indels with end-to-end alignments. The thing OP really needs to do is find out why the alignment rate is so low without doing local alignment.

[carloshgcosta]

Well, telling by experience. End-to-end works well if you want only to look inside a single strain, not consindering SNPs. However, when dealing with metagenomic datasets, you might want to use local alignment(using --sensitive or --very-sensitive), once you might have inside your data more that just one strain of your interest. You could use both, make an assembly, and compare your results.