Usage of fastTopics for multi condition multi timepoints datasets

[JonasDBM]

Hi,

many thanks for developing and maintaining this very useful tool!

I have been successfully using fastTopics for scRNA seq analysis of B and T cell differentiation and it added significantly into insights gained by traditional hard clustering approaches.

A lot of public data nowadays yields informations from multiple samples, conditions and timepoints and I am not sure about how to deal with this data: I usually fit a topic model to the combined dataset and split the structure plots by condition/timepoint. However, I would be curious about your feedback on if there is a more statistically sound way in comparing the topic proportions abundance between conditions of interest.

Many thanks for your help!

[pcarbo]

@JonasDBM It is possible that the topic model might capture sample, condition and/or time-point effects as different topics, but it doesn't always work very well, and it isn't designed for this. Nonetheless I would encourage you to try it; it might produce reasonable results, or it might do something undesirable such as split the "interesting" topics by sample or condition.

An alternative approach you may want to explore described here. It is more complicated than fastTopics, but it is also more flexible, and might better accommodate the challenges of your data set. Also if you have questions about flashier we would be happy to answer them.

[JonasDBM]

Many thanks for your help and pointing out flashier! I will give it a try with both!