Fig. 5: AML Drug Response Analysis.

a Heatmap showing the tile discovered by CHESSBOARD in LSVs from 70 AML related genes (samples = 477, LSVs = 90). The signal (samples = 214, LSVs = 66) is outlined in red. The top “Genome Wide Clustering” track shows sample grouping in Fig. 3a. b Barplot showing for each categorical variable (mutation presence or splicing cluster assignment, left) the drug (right) with the maximum AUC variance explained (x-axis) by the corresponding variable. c The proposed decision tree for administering Sorafenib based on splicing patterns and mutations. Patients with FLT3-ITD- and a signal group splicing pattern exhibit a worse response (high AUC) compared to patients with FLT3-ITD+ and a background group splicing pattern (low AUC). d Violin plots of AUC values for patients' response to Sorafenib when split according to the groups indicated on the x-axis. When combining both splicing and mutation information using the decision tree in (c), the variance explained increases to 36.8%. The bars at the top indicate the total number of samples that fall into each category. Notably, the groups exhibiting favorable drug response (FLT3-ITD+ & Background) are enriched for abnormal splicing (55/66 patients) while the group with poor response (FLT3-ITD− & Signal) are enriched for normal splicing (152/169). Here, abnormal splicing is defined as constitutive expression of the canonical isoform with Ψ1 > 0.9 and Ψ2 > 0.9. e Differential splicing events in FLT3 and EZH2 between the subgroups. For FLT3, the inclusion of exon 4b in LSV1 and exon 17b in LSV2 results in introduction of a frameshift or PTC respectively. Scatterplot (bottom left) shows correlation between Ψ values for the skipping event in FLT3 (Ψ1 for LSV1, Ψ2 for LSV2), while correlation plots (bottom middle and right) show Pearson’s correlation between Ψ and FLT3 expression. The red line indicates the linear regression fit and the band represents the 95% confidence interval. For EZH2, the ΔΨ values between the clusters for these deleterious events in EZH2 are low (<0.2), but are part of a change involving a higher rate of missingness in the background cluster (>0.15).