Unlocking the undruggable spliceosome: generative AI and structural dynamics in cancer therapy

The spliceosome is a dynamic molecular machine essential for transcriptome diversity, yet its complexity creates specific vulnerabilities in cancer. Recurrent somatic mutations in core factors, particularly SF3B1, U2AF1, and SRSF2, drive malignancies by altering splice-site recognition. Such structural perturbations do not merely drive oncogenesis but manifest as distinctive molecular signatures that can serve as potent diagnostic and prognostic biomarkers. However, therapeutic exploitation of these defects remains challenging. This review argues that unlocking the spliceosome requires a shift from static cryo-EM snapshots to dynamic structural ensembles. We explore how physics-based molecular simulation and enhanced sampling methods are merging with generative Artificial Intelligence to identify intermediate states, map cryptic allosteric pockets and target intrinsically disordered regions. Translating these mechanistic insights into the clinic, we evaluate the next-generation of therapeutic strategies, ranging from novel molecular biomarkers to rationally designed allosteric modulators and synthetic lethality. Finally, we discuss how deciphering these altered structural dynamics can guide the identification of splicing-derived neoantigens and biomarkers, establishing a roadmap for precision immunotherapy.