Supervised deconvolution

DS Daniel Spakowicz
SL Shaoke Lou
BB Brian Barron
JG Jose L. Gomez
TL Tianxiao Li
QL Qing Liu
NG Nicole Grant
XY Xiting Yan
RH Rebecca Hoyd
GW George Weinstock
GC Geoffrey L. Chupp
MG Mark Gerstein
request Request a Protocol
ask Ask a question
Favorite

The gene table (G) was deconvolved using the transcriptomes from 22 flow cytometry-sorted and sequenced immune cell types using the CIBERSORT tool [8]. Briefly, characteristic gene expression patterns for each cell type were used to estimate the fraction of each cell type in the sample by solving for the equation:

where G is the gene table of human protein-coding gene expression from the exceRpt pipeline, F is the cell fraction table, and S′ is the reference cell-type gene expression signature from experimentally isolated and sequenced cell types. Support vector regression was used to perform variable selection, reducing the number of characteristic genes used to distinguish cell types and thereby reducing overfitting. The above equation was then solved to provide an estimate of F. P values for the fit of F to G demonstrated that all samples were significant at α = 0.05. Following the solution of F, the cell type expression, S, expanded to all the other, non-signature genes, were inferred using the same equation.

Do you have any questions about this protocol?

Post your question to gather feedback from the community. We will also invite the authors of this article to respond.

post Post a Question
0 Q&A