Hello,
I want to construct a neighbour net on a dataset generated from GBS. (De novo assembly in ipyrad, 130 samples). Depending on how I filter, the dataset is quite large. When I filter more stringently (loci must be present in at least half the samples), the file loads fine in SplitsTree 4.17.0. However, I would ideally like to include more loci. At this point, loading the file becomes extremely slow. (After several hours, nothing seems to happen.)
Does someone have experience with constructing neighbor nets with large data? Is there a better implementation for this? E.g. I know the neighbor net algorithm is part of the R package phangorn (with the note that is experimental). Or would you suggest I decrease the size of my dataset e.g. by choosing one SNP per locus? So far, I am using the nexus files written by ipyrad (the full alignment).
HI Mareike - in my experience it is rather a problem of the performance of your computer than of Splitstree itself. You may either try SplitsTree5, or, as I would suggest to do first, use the ipyrad *.u.snps output file as input for Splitstree. It is anyway better to use unlinked SNPs for NeighbourNET (as well as structure analysis and similar). The u.snps. file is much smaller in size, although you may include many taxa and loci. It might also be a problem that the phylip file is not read by SplitsTree. In this case you can simply transfer the .phy in a .nex by hand (adding the Nexus head and end manually). This should also work. If these suggestions do not help, just let me know and we can solve the problem via PN.
Hi Mareike - I agree to Natascha. The unlinked SNPs file is most appropriate for NNet analysis in SplitsTree as, for example, we did in our paper (Karbstein et al., 2020 TAXON; Article Phylogenomics supported by geometric morphometrics reveals d...
) or as I didArticle Moving beyond assumptions: Polyploidy and environmental effe...
. I used an edited *u.snps.file with ca. 100,000 RADseq loci and 45/280 samples, and had therefore to increase the working memory to 8-10 GB within SplitsTree (can be specified in settings). With 2 GB working memory, SplitsTree usually crashed when analyzing large RAD-seq datasets.
Hello Kevin, hello Natascha,
Sorry for the late reply. I appreciate your help and have calculated the net accordingly. Worked fine :)
Hi Mareike,
another approach (probably too late now) can be to calculate the distance matrix outside of SplitsTree since this is often the most time-consuming step. You can, for example, use the R package ape to calculate a DNA-based distance matrix, save that matrix as Phylip distance file, and finally load that distance file into SplitsTree. This has the additional benefit that you have access to genetic distances that are not implemented in SplitsTree.
Hi Tankred,
Thank you, I think the hint is actually quite helpful. I hadn't considered that and may try to do so in the future.
- Badges
- Science topic
- Similar topics
- Control Systems Engineering
- Filtering