Fig. 4: Tissue-specific features improve the model prediction.

a Left: Distribution of average precision (AUPRC) across 27 GTEx tissue types. P-values were obtained using the paired two-sided Wilcoxon test. AbExp combines various variant and tissue annotations to predict aberrant gene expression and outperforms LOFTEE by about sevenfold. Right: Precision-recall curve for all tissues combined. b Fold change of gene expression against normalized gene expression rank for LTBP3 in tibial artery, an autosomal recessive gene whose defect can lead to dental anomalies and short stature²⁸, and for OR2W3 in blood, an olfactory gene whose defect should not impair the viability of an individual²⁷. Expression outliers are highlighted in red. LTBP3 is tightly regulated with a fold change range of ±20% among non-outliers. The individual marked in red carries a heterozygous frameshift variant that associates with 30% reduction and which is detected as an outlier. In contrast, OR2W3 shows very large variations where individuals with 30% reductions are not outliers. c Distribution of gene expression fold changes for 1295 underexpression outliers (red) and 20,655 non-outliers (gray) harboring a frameshift variant per decile of expression variability (biological coefficient of variation). d Left: Proportion of underexpressed outliers (red; n = 2051), non-outliers (gray; n = 2,749,843), and overexpressed outliers (blue; n = 1177) with a rare variant within a certain distance of the transcription start site (TSS), normalized by the length of each distance interval. Error bars mark 95% binomial confidence intervals. Right: Distribution of Enformer scores in different outlier classes among rare variants within a certain distance of the TSS. For all boxplots: Center line, median; box limits, first and third quartiles; whiskers span all data within 1.5 interquartile ranges of the lower and upper quartiles.