aakrosh
commented
3 years ago The assignments from individual data types will also be useful for interpretation. Current interpretation highlights the features that are most important for each identified class. However, the information from assignments of clusters from individual data types might tell us about the data types that drive each class (rather than the individual features). For example, methylation may be more relevant to why a particular class is being identified in the multi-omic analysis.
The typical use case for sumo is integration of multiple data types. But a user might want to investigate clusters generated from a single data type. In that case, the current formulation which factorizes A_i=HS_iH^T might not be appropriate. A formulation A=HH^T similar to that used in SymNMF (https://link.springer.com/article/10.1007/s10898-014-0247-2) might be a better choice for a single data type.