Abstract B002: Computational Discovery of Evolutionary Synthetic Lethality Networks for Personalized Gene Therapy and Small-Molecule Design in Treatment-Resistant Pediatric High-Grade Glioma

Abstract

Pediatric high-grade gliomas (pHGGs), including diffuse midline glioma (DMG), remain among the most lethal childhood malignancies, with median survival typically below 18 months despite maximal surgery, radiotherapy, and chemotherapy. Disease progression is driven by highly plastic tumor cell populations that survive treatment. We developed an integrative computational framework to reconstruct resistance evolution, identify state-specific therapeutic vulnerabilities, and design personalized gene therapy and blood-brain barrier (BBB)-penetrant small-molecule therapeutics targeting treatment-resistant pediatric glioma. Single-cell RNA sequencing data from m cells across 74 pediatric high-grade glioma specimens were integrated with whole-exome sequencing, DNA methylation, spatial transcriptomics, and clinical outcome datasets. RNA velocity, pseudotime inference, and graph-based lineage reconstruction were used to model therapy-induced transitions among proliferative, oligodendrocyte precursor-like, mesenchymal-like, stem-like, and treatment-persistent cellular states. Dynamic gene regulatory networks were reconstructed using transcription factor activity inference and protein interaction networks to identify resistance-associated signaling modules. Candidate evolutionary synthetic lethal interactions were prioritized by integrating genome-wide CRISPR dependency datasets, conserved developmental pathways, network centrality metrics, and predicted essentiality scores. Personalized CRISPR interference and siRNA therapeutic candidates were computationally optimized using guide efficiency, off-target prediction, and genomic specificity analyses. Parallel chemoinformatic drug discovery employed graph neural network–guided molecular generation, structure-based virtual screening, molecular docking, molecular dynamics simulations, and multi-objective ADMET optimization to identify BBB-permeable compounds targeting the same adaptive resistance proteins. Trajectory identified multiple convergent resistance pathways leading to persistent mesenchymal-like and stem-like cellular states following treatment exposure. Network analysis predicted 37 high-confidence evolutionary synthetic lethal interactions, with enrichment in chromatin remodeling, DNA damage repair, developmental signaling, mitochondrial metabolism, and stress-response pathways. Prioritization algorithms identified 11 candidate gene therapy targets predicted to exhibit high state specificity while minimizing dependency in normal developing neural cells. Chemoinformatic screening evaluated approximately 2.1 million virtual compounds, producing 42 lead molecules satisfying BBB permeability, synthetic accessibility, predicted low toxicity, and favorable docking against prioritized resistance-associated proteins. Multi-objective ranking integrated binding affinity, molecular dynamics stability, predicted pharmacokinetic properties, and target selectivity to generate individualized therapeutic combinations. Shivi Kumar, Philip Moheno, Sweta Gupta. Computational Discovery of Evolutionary Synthetic Lethality Networks for Personalized Gene Therapy and Small-Molecule Design in Treatment-Resistant Pediatric High-Grade Glioma [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Bridging Discovery and Clinical Impact in Pediatric Cancer; 2026 Sep 22-25; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2026;86(18_Suppl_1):Abstract nr B002.