Skip to main content

Table 1 Description of each baseline method included in the benchmark

From: DECONbench: a benchmarking platform dedicated to deconvolution methods for tumor heterogeneity quantification

Name RNA_wICA RNA_wNMF DNAm_EDec DNAm_MeDeCom DNAm_wICA both_wICA both_wNMFMeDeCom both_meanwNMFMeDeCom
Acronym r_WIC r_WNM m_EDC m_MDC m_WIC b_WIC b_COM b_MEA
Data type RNA RNA DNAm DNAm DNAm both both both
Feature Selection DNAm / / 5.000 most variable probes 5.000 most variable probes 5.000 most variable probes 5.000 most variable probes 5.000 most variable probes 5.000 most variable probes
Feature Selection RNA ICA, selection of top-contributing genes and filtering of duplicated genes ICA, selection of top-contributing genes / / / / ICA, selection of top-contributing genes ICA, selection of top-contributing genes
Deconvolution algorithm DNAm / / Edec MeDeCom ICA weighted by top-contributing probes ICA weighted by top-contributing probes MeDeCom with the A matrix computed on RNA as startA parameter MeDeCom
Deconvolution algorithm RNA ICA weighted by top-contributing genes NMF with snmf/r method / / / ICA weighted by top-contributing genes NMF with snmf/r method NMF with snmf/r method
Multi-omic integration strategy / / / / / Averaged DNAm and RNAm proportion matrix DNAm deconvolution uses RNA deconvolution as input Averaged DNAm and RNAm proportion matrix
Time 10 A ~ 10 min ~ 20 min ~ 3 h ~ 17 h ~ 10 min ~ 10 min ~ 17 h ~ 17 h 30 min
Time 1 A ~ 1 min ~ 2 min ~ 20 min ~ 1 h 40 ~ 1 min ~ 1 min ~ 1 h 40 min ~ 1 h 45 min
Reference of the tools/algorithms used Hyvarinen [25] Frichot et al. [26] Onuchic et al. [9] Lutsik et al. [8] Hyvarinen [25] Hyvarinen [25] Lutsik et al. [8] and Frichot et al. [26] Lutsik et al. [8] and Frichot et al. [26]
  1. A baseline method is composed of two steps: [1] feature selection and [2] deconvolution algorithm. All deconvolution algorithms used as baseline are already published and documented in the literature (see Reference of the tools/algorithms used). A detailed description of the coding instruction and a mathematical description of the algorithms can be found in the "Methods" section. Source code is publicly available on the DECONbench platform. Time 10 A corresponds to the approximated computation time to estimate 10 proportion matrices A (corresponding to the test sets hidden on the platform). Time 1 A corresponds to the approximated computation time to estimate 1 proportion matrix A (closer to real applications on one dataset)
  2. Bold acrononyms are used to identify methods in Figs. 3 and 4