I'm trying to detect infiltration of immune cells in RNAseq data (TCGA_LUAD), In the beginning, I performed cibersortx (through immunedeconv R pavkage) on 100 samples. But when I change the sample number from 100 to 200, I do observe a change in score from the two runs/process.
For example; Sample_A shows maximum infiltration for Macrophages_M2 and Neutrophils in RUN_1. The same sample (Sample_A) will show a higher score/infiltration for Tcells_CD8 or Bcells in RUN_2.
Why do we see such a difference in results? Is this because of the deconvolution method? If yes, then how can we normalize the parameters or gene counts to get more redundancy across the results? So that I can select the right samples showing infiltration in particular cell types!