Supplementary Figure 6: cellAlign performance under noise and pre-processing methodologies.

(a) The effect of increasing levels of noise in the form of additional dropout events on local alignment. Assessment was performed by calculating the fraction of the originally identified, locally-conserved region along the PAM trajectory that was still identified in the noised data. Error bars and centers denote standard deviations and mean values across n=500 simulations. (b) Performance of cellAlign for data processed by different preprocessing techniques, including: cells subsampling (purple), normalization method (turquoise), trajectory building algorithms (red) and application of data-imputation (green). Spearman correlation of alignment-based distances calculated across gene modules between original and the modified expression data are shown. For cells subsampling, the average and standard deviation across ten simulations are shown as error bars. (c) Spearman correlations between alignment based distances calculated across modules of co-expressed genes using expression data obtained either without batch correction or corrected by Combat or Limma.