Overall paper outline

Overall Key Points

• Multi-modal data is everywhere nowadays
• Tensor-form analysis has improvements in data reduction, modeling, identifying mode-specific effects, and integration
• Maybe introduce tensorpack/tensordata as some useful tools for tensor methods in biology?

Figures

Figure 1

• Cartoon of overall approach
• What is multi-dimensional vs. multi-modal
• Something to introduce the various datasets we'll apply

Figure 2: Tensor methods reduce data down better

• R2X vs. size heavily prominent here
• It may be helpful to introduce other forms, like Tucker, here
• Maybe there is a way to visually depict that tensor methods operate by reducing duplication in the scores/loadings?

Figure 3: Better at separating mode-specific effects

• MEMA separation
• Simulation?

Figure 4: Mode-specific separation allows for data integration

• It would be good to first discuss what are the best opportunities for integration
• Maybe MEMA across cell lines?
• Linking LINCS datasets

Figure 5: Better at modeling the data (shown through imputation)

• Chord-wise imputation
• Describe improvements to make structured missingness work better

Figure 6: Better at sparse sampling

• Random position imputation, up to very high missingness
• I think we can claim to do better than others here?