Questions on Use Cases

[Olaf Baumeister Warren Holmes Corning]

Hi all,

Is it appropriate to use Garnett to separate cells that are highly expressing gene X and lowly expressing gene Y, inverse, or highly expressing both? The ability to integrate other gene features and create its own expression thresholds based on the data seems preferential over a boolean subsetting. 

How much stock do you put into the ambiguity warning?

Best,

Olaf 

[hpl...@gmail.com]

Hello, Garnett isn't ideally set up for that case, but in practice it usually does ok because a cell is assigned a type if the 'expressed' marker gene is in the 75th percentile of expression - so very low expression is generally excluded. The ambiguity warning will depend a lot on the dataset - you can see in practice whether it's a problem by how many 'Unknown' type cells you end up with after classifying.