number of Illumina reads for transcriptome assembly

[glopatriello]

Hi there!

I am performing a de novo transcriptome assembly of a non model organism. To do that, we have sequenced RNA with ONT and Illumina and we are going to build the transcriptome using the hybrid approach.

I was wondering if in the hybrid approach rnaSPAdes requires a certain number of sequenced Illumina reads(for istance 200 milion of fragments) or the more I sequence the better the results will be.

Best regards,
Giulia

[andrewprzh]

Dear Giulia

There is no certain limit for number of short reads you provide. However, low-covered datasets (let's say below 10-15 M reads may give sub-optimal results). Very high coverage datasets, on the other hand, may take a long time to assemble. From our experience 50-200M reads is a good balance for typical eukaryotic transcriptome.

Best
Andrey

[glopatriello]

Thank you Andrey for the fast answer!

I was wondering if we have sequenced more than 200 M of fragments ,for example 1B of fragments, if we use the entire datasets could introduce some noise into the transcriptome assembly?
In addition, to perform the sub-sampling of the reads do you suggest to perform just a sub-sampling in terms of number of input reads or it is better to use the reads normalization approach?

Best,
Giulia

[andrewprzh]

I cannot say for sure as we never assembled 1B reads. But yes, extremely high coverage datasets may possibly result in some excessive transcript sequences.

As for down-sampling - I personally used usual sub-sampling (random), but some users tried normalization approach as well. Not sure about the results, I've some controversial feedback on normalization procedures.
In any case, rnaSPAdes is designed to handle non-uniform coverage and usual sub-sampling should work just fine (and probably faster than normalization). Thus, I'd stick to the simple one.

Best
Andrey