Extended Data Fig. 4: Overall score of integration including different weightings of bioconservation and batch correction.
From: Modeling fragment counts improves single-cell ATAC-seq analysis
a) Comparison of integration accuracy for embeddings generated with Poisson VAE, Binary VAE, PeakVI, Signac, cisTopic, and SCALE on the four datasets. For cisTopic and Signac, additional batch correction was performed using Harmony. Overall integration accuracy scores were computed using a 40:60-weighted mean of batch correction and bioconservation scores. P values were computed using the two-sided paired Wilcoxon test; Benjamini–Hochberg corrected. Error bars represent the 95% confidence interval over ten cross-validation runs. b) Overall score computed from different bioconservation and batch correction weightings.
