closely related species as a reference for seed

[shri1984]

Hi, I am working on a genome expected size is around 10g. I just wonder is it possible to use a closely related genome as a seed for the alignment?

[MarkusRainerSchmidt]

Hi,

Generally speaking, it is possible to align against any genome you want. The less related the genome, the more the aligner will struggle to produce good alignments. By adjusting parameters, you should be able to offset this to a certain degree. If you think that you are getting suboptimal alignments, I would suggest trying this:

• Switch to SMEMs -s SMEMs. SMEMs are a seed set that contains much more seeds that the max. spanning seed set; computing them will make the aligner slower but you may get more hits.
• Reduce drop-offs as e.g. --SoC_Score_Drop-off 0.05 and --Seeding_Drop-off_B_-_Factor 0.0025. This will make the aligner look at more candidates for alignments – if such candidates are produced by the seeds.

It looks like you are working with a large genome. Genome size by itself is not a problem, however if the reference genome you are using contains many repeats, it might be necessary to increase the minimal seed length: e.g. -l 17/-l 18. This parameter has an upper and a lower limit:

• If set so too low, millions of “noise seeds” will be computed and lead to slow and wrong alignments. This minimal value is determined by the repetitiveness of the genome you use as reference. E.g. if 90% of all 16nt long sequences exist >100.000 times on your genome a minimal seed size of 15nt is too low. The --Maximal_Ambiguitycontrols at what point seeds are discarded; increasing this value will extend the minimal seed length range downwards for the price of additional runtime.
• If set too high no seeds and therefore no alignments will be computed. This maximal value is determined by the difference between read and genome. E.g. if the read has an error every 15nt a minimal seed size of 16nt is too high.

Here, your specific combination of a large genome (and therefore presumably repetitive genome) that is only form a related species might mean that there is very few wiggle room for this parameter.

By using the python interface, it is possible to check the computed seeds and SoCs (before as well as after the harmonization). So if you can’t get a good parameter combination by trial and error this is the way to look into what’s going on inside the aligner (if you need help with this just let us know).

[shri1984]

Thank you very much. I am on it now. Let you know how it goes.