Hello, I join the channel because I will be using multiple of the pipelines that are involve in this meta-omics general pipeline but I was wondering if this channel can also be use to discuss more of general questions that I have in regards of what approach to take? I have environmental samples from the Southern ocean and I dont know if I should do single assemblies for all of them or to do coassembly with the ones that have similar communities. I just found the #C031QH57DSS pipeline and it might help me to decided it but I have already run the mTAGs and simka which are also use to search for the disimilarities and I still cant make a decision. What do you think as developers? Is it worth to do coassembly or not? what does it depends on? Also, how do you make the clustering for the coassembly? Because if you use a tool that uses databases, you are adding a lot of bias to your decision no? Sorry if this is not the place to discuss something like this, I'm just stuck and need community knowledge to make a decision...

I assume you're thinking of coassembly for MAG reconstruction. In theory, I prefer to do that to get a single set of MAGs but with today's enormous sequence output it seldom works and I have to resort to per sample assembly followed by dereplication (not yet in #CE9MS66BS, I use dRep). Another possible disadvantage of coassembly is of course that the assembly might be too complicated and result in certain members of communities not fully present in the final result. An in between would be to, as you allude to, do group assembly after running #C031QH57DSS to understand which samples are more similar to each other. I wouldn't worry too much about the database bias if I used this just to group your samples for assembly. And again, the risk is high that you will have to run single-sample assembly anyway.

From my limited knowledge of coassembly it depends on the similarity of the samples (which if they are across the entire southern ocean, and at different depths maybe there could be quite variation) and also 'how good' you want tyour assemblies to be. If you don't mind more fragmented assemblies (you only care about recovering sufficiently long contigs for detecting particualr genes, maybe it doesn't matteR)

So the samples are all from a specific area around the kerguelen islands but we have different types of stations, like ones more in the bay area, other in lakes, others off shore... i wouldnt expect all of them to be the same but some of them should be quite similar. My thoughts were that doing coassembly is supposed to improve the resulted MAGs but if Im going to do simple assemblies to validate the coassembly, I dont really see the advantage in doing coassembly. Is it possible in #CE9MS66BS to do single assembly for all the samples and then do the binning with the samples that are taxonomically more similar? would that be a better a approach to resolving the possible fictional members of the community?

I would definitely agree that it depends on the level of similarity among samples. If two samples are completely dissimilar and share no sequence, there is no advantage to coassembly at all. By comparison, to samples that contain identical sequence might benefit by increasing coverage of low-coverage taxa or or hard-to-assemble regions. It's the in-between - where most real samples fall - where the issue turns a bit more thorny. Ultimately, coassembly might improve the assembly quality, particularly for low-coverage taxa, by sharing sequence information across samples. However, you are more likely to end up with chimeric sequence as similar, but not identical, genomes will end up being conflated in the assembly graph, and it can increase fragmentation of the output. Thus, I tend to think co-assembly is best when you either have a specific target you are looking to assemble, or you know the samples are likely to share a significant proportion of genetic material (e.g. replicates from the same location). Otherwise, I would tend to recommend single-sample assembly as it both a) takes less resources, and b) will likely give you more accurate results

Thank you very much for your answers! They have been really helpful to decide thank you!!