Extended Data Fig. 1: Generalization performance of XPert and baseline models across cell lines under the cold-cell setting.
(a) Boxplots of predictive metrics (MSE, Pearson (xpert), and Pearson (xdeg)) for each cell line on the L1000_sdst dataset. Individual cell lines are shown as colored scatter points (N = 164). The central line within each box denotes the median; the box limits represent the interquartile range (IQR, from the 25th to 75th percentile); whiskers extend to 1.5× IQR beyond the box limits. (b) Scatter plot of the maximum Spearman correlation between test and training cell lines versus the predictive performance (Pearson (deg)). The gray line denotes the linear regression fit. (c) Distribution of maximum Spearman correlation values across test cell lines, with group boundaries defined by mean ± standard deviation (purple dashed lines). Cell lines were stratified into Low (N = 22), Medium (N = 110), and High (N = 31) similarity groups. (d) Box-and-whisker plots comparing the prediction performance distribution (MSE, left; Pearson (deg), right) across the three cell similarity groups (Low (N = 22), Medium (N = 110), and High (N = 31)). The central line within each box denotes the median; the box limits represent the interquartile range (IQR, from the 25th to 75th percentile); whiskers extend to 1.5× IQR beyond the box limits; and outliers are shown as individual points beyond the whiskers.
