Can MELD be applied to time series data? Is this an appropriate approach? #68
Description
Hi, first of all, thank you for developing MELD! We’ve successfully applied it to some of our datasets, and the tutorials were clear and very helpful.
I have a question about extending MELD to time series data, specifically in the context of a regeneration study. I apologize if this is a basic or silly question or if there are gaps in my understanding, but I’m curious about whether MELD is appropriate for analyzing time-series data where each time point (some n-timepoint) would be treated as a separate condition.
In our data/experiment, we assume that there is some overlap in cell types/states between these time points. As regeneration progresses, we expect the distribution of cell states to shift from one time point to another, with some cell populations being more affected by the process than others.
Given this, I’m wondering:
- Is it appropriate to apply MELD in this time-series context at ALL or should we look at something else?
- What considerations should be made when using MELD for such an analysis?
- Are there specific caveats or best practices when interpreting the likelihood scores across multiple time points (i recall that in one of the figures there was also an experiment where there were multiple conditions all at the same time? should the results be at all interpreted in that same manner)
Thank you in advance for your insights. I look forward to your input!