Hello BioStars community,

I am trying to integrate a Jurkat WT/KO multi-omics dataset using the MOFA2 R package.

The dataset consists of three views:

1. RNA-seq (WT and KO replicates)
2. Proteomics (WT and KO replicates)
3. Phosphoproteomics (measured from the exact same lysates as the proteomics samples)

Here is the exact matching nature of my experimental setup:

• Proteomics and phosphoproteomics are completely sample-matched (derived from the same lysates).
• RNA-seq was generated in a separate, independent experiment and is NOT matched to the proteomics/phosphoproteomics samples.
• The RNA samples represent the exact same biological conditions (WT or KO) but do not originate from the same biological replicates as the protein/phospho samples.

Conceptually, the design looks like this:

Protein sample P1 <-> Phospho sample P1 (matched) RNA sample R1 [not equal to] Protein sample P1 (unmatched, same condition only)

My understanding is that standard MOFA2 assumes that samples correspond to the exact same biological entity across all views to calculate covariance.

Given this setup, I have three questions:

1. Is standard MOFA2 appropriate at all for this specific configuration?
2. If so, is there any normalisation or other techniques with which I can still perform MOFA2?
3. Is this design fundamentally incompatible with standard MOFA2? If so, are there multi-group extensions or alternative frameworks (like mixOmics/DIABLO or specific matrix factorization variants) that can anchor the integration using the experimental group design matrix (WT vs KO) rather than strict sample-to-sample matching?

I would highly appreciate guidance or code architecture examples from anyone who has successfully worked with partially matched or uncoupled multi-omics datasets in factor-analysis frameworks.

Thank you!