Fig. 1: Concept and design.

a Differential SJ Patterns due to AKR1C2 overexpression by constitutive activation of the KEAP1-NRF2 System. Sashimi plot illustrating the impact of AKR1C2 overexpression on SJ dynamics. The genomic coordinates are displayed along the x-axis, while the y-axis indicates the read counts. The thickness of the lines represents the number of reads supporting the junction. The red section represents a control sample with no KEAP1-NRF2 system alterations, exhibiting a standard junction pattern. Contrastingly, the sample affected by KEAP1-NRF2 system disturbance shows both a pronounced AKR1C2 overexpression, evidenced by increased SJ counts, and the presence of abnormal SJ at previously unannotated locations. b Structure of the pipeline. Our model is first designed using only Active and Inactive samples, setting Uncertain samples aside for posterior analysis. From this dataset, abnormal SJs and their corresponding normal SJs were selected as the input required for the construction and evaluation of the model generated. With the model built, we evaluated and integrated the Uncertain mutations, leading to the rebuild of the model with an updated abnormal SJ pair selection. An initial cross-validation analysis was performed, and then the complete model using the relabeled dataset was generated and implemented on the SRA dataset. c Abnormal SJ selection. For each cancer type, abnormal SJs were evaluated using a one-tailed Wilcoxon test to compare their presence between Active and Inactive samples. The p-values were adjusted using the Bonferroni method. Abnormal SJs with adjusted p-value ≤ 0.001 were selected from each cancer type and compiled as inputs for our classifier.