Hi everyone,
I’m working on deconvolution of bulk RNA-seq data using CIBERSORT with the LM22 signature matrix to estimate immune cell proportions. My supervisor asked a very valid and challenging question that I’d like to get your input on:
How believable or reliable are the output proportions from CIBERSORT, especially for rare cell types like GD?
In my current analysis, I’ve followed a methodology similar to the one used in:
A Characterization of the immune cell landscape in CRC: Clinical implications of tumour-infiltrating leukocytes in early- and late-stage CRC
However, I also came across this recent tool and paper:
ReCIDE (Robust Estimation of Cell Type Proportions by Integrating Single-Reference-Based Deconvolutions), which emphasizes that rare cell type estimates (typically <2%) may not be reliable when using single-reference-based methods like CIBERSORT.
This leads me to a few key questions I would really appreciate feedback on:
How much confidence can we have in CIBERSORT results for rare cell types (e.g., <2% estimated proportion)?
Are there better-suited approaches or additional validation steps for rare cell types in bulk RNA-seq deconvolution?
Has anyone tried integrating CIBERSORT with other methods like ReCIDE or using single-cell data to improve deconvolution accuracy for rare populations?
I’d really value hearing how others in the field handle these limitations, especially when trying to interpret or validate rare cell population estimates from deconvolution analyses.
Thanks in advance!
