Trinity Multi-species compatibility

Question

I am using Trinity from last couple-up months or so. Initially, using same species in different condition was worked fine for me. Now i am trying to do a DE analysis using multiple species. Actually, i have 3 fish performs similar behaviour and another 3 performs similar strategy. Here i considered 3 similar species as replicate for one type (and totally i have two type) . Though these are 6 different species but belong to same family.

I have already made a combine assembly according to trinity protocol (--samples_file). After assembly i have checked the separate alignment rate for each species and it was above 85%. And lastly i checked heat-map and it gave me a result that somehow match with my expectation.

But i am a bit confused about the possible generation of "assembly chimera". In most of the cases i found people do DE analysis with same species. Also didn't find any reference article that did multiple species DE analysis using Trinity pipeline. So it will be a great help for me if i get some suggestion that either i should go with existing result or i need to find another way.

Answer 1

Given that you're using multiple species / samples, the results from this are going to be quite weird.

For doing such a comparison, I think the best approach is to do a combined assembly (as it looks like you've done) to generate a single chimeric transcriptome (which will not be a good representation of any of the component species), then do a differential expression analysis using that combined assembly as the reference.

Ideally you would have multiple replicates for each different species, and then add the species and conditions into the statistical model, so that any species-specific variation can be identified and/or filtered out.

Six separate assemblies that are later merged will be less useful for a differential expression, because it will bias towards differences in the assemblies. This is a similar rationale to that used for generating UMAPs from multiple single-cell datasets where a combined UMAP is generated for the full dataset with identical parameters, so that sample-specific differences in the UMAP parameters are less of an issue.

I think it will be okay to go ahead with the combined assembly, bearing in mind that the lack of replicates for the differential expression study is going to make it more difficult to identify biological significance. Using split assemblies will not make that better; if anything, it'll make it worse.