Open artur-sannikov opened 1 year ago
We can certainly add another MAE demo data set in mia, for instance. It should be about microbiome research (which is indeed so far less covered in terms of multiomics methodology than cancer studies).
Or we can use existing data set. The possible sources:
I expect that more informative factors can be identified from data sets with larger sample sizes.
Is your feature request related to a problem? Please describe. At the moment, Hintikka data is used for basically all analysis in OMA book. However, I see a problem in the MOFA section because the model only finds one factor which explains the variability only in metabolomic data (see "Variance Explained per factor and assay" figures). So I have difficulties interpreting and discussing the results because it does not show much in my opinion.
In contrast, in the original MOFA+ paper, they found that the factors capture different pieces of information, for example the differences in methylation, classes of neurons, etc. The presence of these factors also allowed them to apply t-SNE to discover sub-populations of cell types. Well, in our case, we cannot do much of downstream analysis.
Describe the solution you'd like I see two solutions here:
These two solutions can be implemented simultaneously, and I do not have any preference to either as long as the data provides us with meaningful and interpretable results.
Additional context