search

navinlabcode / copykat

Other
193 stars 54 forks source link

Parameter UP.DR and parameter LOW.DR #21

Closed shangyf-stu closed 3 years ago

shangyf-stu commented 3 years ago

Hi, sir Thanks for your useful tools to prediction of tumor cell. I am adjusting its parameters hoping to get good results. But I could not understand the parameter LOW.DRand UP.DR well. I know they are used to filter genes, but I don’t know how they work. In the help documentation of the R package, you says: LOW.DR: minimal population fractions of genes for smoothing. UP.DR: minimal population fractions of genes for segmentation. Could you give me a more clearly expaination? Especially for UP.DR. And why genes that are expressed in LOW.DR to UP.DR fractions of cells are kept?

best, Shang

gaobio commented 3 years ago

Hi, sir Thanks for your useful tools to prediction of tumor cell. I am adjusting its parameters hoping to get good results. But I could not understand the parameter LOW.DRand UP.DR well. I know they are used to filter genes, but I don’t know how they work. In the help documentation of the R package, you says: LOW.DR: minimal population fractions of genes for smoothing. UP.DR: minimal population fractions of genes for segmentation. Could you give me a more clearly expaination? Especially for UP.DR. And why genes that are expressed in LOW.DR to UP.DR fractions of cells are kept?

best, Shang

Yes, we are trying to filter genes in two ways. In the smoothing step, we tried to retain as many genes as possible, i.e. LOW.DR, which is the lowest cutoff of genes detection rates (fraction of cells that expressed certain genes). In the segmentation step, we tried to use less genes that have higher detection rates. Only genes that are expressed in more than UP.DR cells are used for segmentation. Conceptually, it is another bottom cutoff. We named it as UP.DR to differentiate from LOW.DR. UP.DR should be set to be higher than LOW.DR. Great idea to play around with these parameters.