Fig. 3
From: TOAST: improving reference-free cell composition estimation by cross-cell type differential analysis
TOAST is stable with different initial feature selections in gene expression simulation studies. The panels from top to bottom correspond to sample sizes 50, 100, and 200. The panels from left to right correspond to different methods of selecting initial features: Top 1–1000 variables is to select the top 1000 most variable features, 1001–2000 is to select the top 1001–2000 most variable features, similarly 2001–3000, 5001–6000, 10001–11000 are to select the top 2001–3000, 5001–6000, 10001–11000 most variable features. Random 1000 is to randomly select 1000 features as initial features. In each panel, “Initial” and “TOAST” correspond to reference-free deconvolution results without and with TOAST. The presented results are summarized over 100 Monte Carlo experiments