Hello everyone, I am working on a non-model plant species whose RNA-seq data were generated a few years ago by another group, and I have access to both the raw reads and the previously assembled transcriptome submitted to TSA. I now want to mine ORFs for downstream analyses, and I am trying to determine whether it is better to rely on the existing assembly or to generate a new one from the raw data, especially considering that current assembly pipelines have improved since the original work. I would also like to know if there are any reasons, beyond the availability of updated tools, that would justify creating a new assembly. I have no computational limitations and would appreciate any guidance or experience that can help inform this decision. Thank you.
There are a few factors to consider:
The quality of the existing reference.
If there is a high quality chromosome level assembly, you would need to be skilled and have good data to replicate this quality. And this is no easy feat even with modern tools, especially if your plant has a complex genome (e.g., high levels of ploidy). If the existing assembly is highly fragmented, creating a similar quality one is considerably less work. I realise this will be a transcriptome assembly, but in my experience transcriptomes derived from genome assemblies are better - better ability to discern between isoforms and duplications, for example.
Evolutionary distance between the sample populations and the reference.
Research has shown that mapping efficiency and downstream inferences are significantly affected by evolutionary distance from population to reference (for clarity, I am an author on that paper). If your samples are significantly different from the current reference, it may be worth generating a new one, but you'd have to measure the work to improvement tradeoff.
Sequencing strategy of these "old" samples.
To generate a new good quality transcriptome assembly, especially if you only have RNAseq reads, you'd need high quality long read sequencing.
Thank you for the detailed explanation.
In my case I am not generating new data, so I am limited to the original RNA-seq and the published de novo transcriptome. There is also no genome assembly available yet. Since the original study used standard tools, evaluated the assembly with mapping rates, BUSCO, and functional annotation, and the results look reasonably good, I’ve decided to proceed with the existing transcriptome assembly as my reference.
I have access to both the raw reads and the previously assembled transcriptome submitted to TSA.
and
I am trying to determine whether it is better to rely on the existing assembly or to generate a new one from the raw data, especially considering that current assembly pipelines have improved since the original work.
Depends on how old the original transcriptome is and what strategy was used to create the assembly. If the assembly was created using one of the standard programs (e.g. trinity) then you are not likely to get drastically better results without using new/additional data (unless the original work was not done to a certain standard).
I have no computational limitations
If that extends to include time you will spend on redoing the assembly then go ahead and try reassembly. You can then compare the results of old with new and see if you manage to get some improvement.
If the genome sequence has since become available then it may be an additional avenue to directly try and predict genes from it to create models that you can use to compare your existing transcriptome to.
Thanks a lot for your clear summary.
In my situation there is no genome assembly available, and I do not plan to add new sequencing, so I am working only with the original data and the published transcriptome. The paper shows that the authors used a standard pipeline and assessed quality with mapping, BUSCO, and annotation, which gives me reasonable confidence in the existing assembly. Based on your advice, I’ll continue with that as my reference for now.
Log in to answer this question.
