Fig. 4: ESM1b predictions in clinically relevant genes depend on the isoform context.
From: Genome-wide prediction of disease variant effects with a deep protein language model
a, The consequences of variants (for example, damaging versus neutral) can depend on the isoform context. b, Comparison of the primary and one of the alternative isoforms of P53. Three specific variants are detailed. c, Left: all 3,477 ClinVar variants with highly variable ESM1b effect scores across different isoforms (defined by s.d. > 2). Center: the lowest and highest isoform scores predicted for all VUS from the left panel (top two boxes), compared to the mean scores (across isoforms) of VUS, benign or pathogenic variants (as in Fig. 2d; bottom three boxes). The boxes represent the Q1–Q3 range and median (Q2) line; whiskers correspond to 1.5× IQR; outliers (outside the whiskers) are shown individually. Right: the distribution of the lowest and highest isoform scores predicted for all VUS from the left panel, compared to the distributions for pathogenic or benign variants from ClinVar, HGMD and gnomAD (as in Fig. 2a). Across all panels, the number of variants associated with each category is shown in parentheses. d, The top 100 ClinVar genes with the highest number of variants with highly variable effect scores (as in c). Numbers of annotated isoforms of each gene are shown in parentheses.
