open syncmer instead of minimzer?

[jianshu93]

Hi @eric9n,

I open an additional issue here about using a different sketching algorithm, in addition to minimizer. I attached slide explaining the differences between minimizer and open syncmer and also the paper on open syncmer (https://peerj.com/articles/10805/).

Ideally, the syncmer/minimizer extracted, are just kmer from the genome and this will not affect how kmers will be compressed in a compacted hash table, just what kmers from genomes will be extracted is different.

The implementation should be straitforward. and I will point to some existing ones. Let me know your thoughts on the open syncmer idea.

Thanks,

Jianshu open_syncmer_kun_peng.pdf

[eric9n]

I think it might have advantages for specific databases, but it would be difficult to make it universally applicable.

[jianshu93]

Actually open syncmer has been proved theoretically to be efficient in the case for reads mapping, e.g., in minimap2 for all cases (if open syncmers are chosen as seeds/anchors and then chain the seed and extend), I attached a paper where the prove based on minimizer is extremely difficult to obtain while for open syncmer it is much easier: https://genome.cshlp.org/content/33/7/1175.short

It seems for other genomes, human for example, it is slower due to repeated kmers, but for bacteria/archaea/virus genomes, not a problem all since those genomes are not multi-copy.

Thanks,

Jianshu

[eric9n]

In the paper you provided, the s-mer lengths for syncmers are generally below 10. Mathematically, this translates to about 2^22 (assuming a 4-letter alphabet), which wouldn't allow for the construction of a very large database.

Different methods are needed to build the database. This approach is not suitable for building a database for Kun Peng.