search

GIS-SP-Group / RCA

R package for robust clustering of single cell RNA sequencing data
MIT License
38 stars 24 forks source link

Cutoff for two global panels #3

Open ahy1221 opened 7 years ago

ahy1221 commented 7 years ago

Hi I am looking at the source code of RCA to learn how the algorithm works. In the featureConstruct() function there is a line 38 data3[data3<=(sysdata$at)[i]] = (sysdata$at)[i] assigning the expression value in globalpanel which smaller than 1.5 or 0.75. I am wondering why we should do this cutoff, would that impact the clustering result ?

Thanks

GIS-SP-Group commented 7 years ago

Hi, ahy1221

FYI, the consideration is the detection limit of micro arrays. The small values at the tale of the distribution (per sample) after normalization tends to be purely noise. The values 1.5 and 0.75 have thus been chosen based on the size of the lower-end peak. Drawing a violin plot for a reference sample would make things clear.

Huipeng