Skip to main content

Table 1 Summary of different feature selection techniques used by reference-free deconvolution methods in published studies

From: TOAST: improving reference-free cell composition estimation by cross-cell type differential analysis

Features selected by RF methods Published studies
  Deconf Liebner et al. [40]
  RefFreeEWAS Johnson et al. [9]
  RefFreeEWAS Johnson et al. [10]
Largest variability RefFreeEWAS Chen et al. [11]
  RefFreeEWAS Everson et al. [41]
  ReFACTor Kaushal et al. [42]
  ReFACTor Rahmani et al. [23]
  NMF Feng et al. [32]
External information Deconf Gaujoux et al. [43]
  RefFreeEWAS Gasparoni et al. [44]
  1. RefFreeEWAS is EWAS using Reference-Free DNA Methylation Mixture Deconvolution, from CRAN package RefFreeEWAS. ReFACTor is reference-free adjustment for cell type composition, from the GLINT package. NMF is non-negative matrix factorization, available from https://github.com/haoharryfeng/cfDNAmethy. Deconf is the in-silico deconfounding approach, i.e., alternate least-square NMF method using heuristic constraints, available from the CellMix package