search

theislab / scib

Benchmarking analysis of data integration tools
MIT License
283 stars 60 forks source link

Questions about the weighted score design of scIB #391

Open HelloWorldLTY opened 7 months ago

HelloWorldLTY commented 7 months ago

Hi, I have a quick question about the setting of the weighted sum: image

I understand to assgin S_bio with 0.6 and S_batch as 0.4 are to ensure bio convservation is more important. However, I wonder what is the motivation for choosing such weight combination. Shall we choose S_bio has weight as 0.7, for example, for some datasets or some tasks? Do we need to perform grid search for this weight in our practical applications? Thanks.

mumichae commented 3 months ago

Hi, this is a very good question. We somewhat arbitrarily chose the weighting to ensure that bio metrics have more importance than the batch metrics, however you might want to try out different weightings , depending on the number of metrics you use and the biological question at hand. @LuckyMD what are your opinions on this?