Abstract B004: Targeting Transient Cell-State Dependencies in Glioblastoma Through Single-Cell Evolutionary Modeling and Structure-Based Drug Design

Abstract

Glioblastoma kills within a median of 15 months even after resection, radiotherapy, and temozolomide, largely because tumor cells adapt to therapy through transient drug-tolerant states long before stable genetic resistance takes hold. Most computational drug discovery pipelines are built on static molecular snapshots and consequently miss these short-lived, functionally critical windows of vulnerability. We present a computational framework that reconstructs the evolutionary trajectory of glioblastoma cells under treatment pressure and uses that trajectory to identify druggable targets that exist only transiently, during state transition. We integrated single-cell RNA-seq data from over 150,000 tumor cells across primary and recurrent glioblastoma specimens, applying batch correction and RNA velocity–based pseudotime inference to map transitions among proneural, classical, mesenchymal, and glioma stem-like states as tumors adapt to treatment. From this, we built dynamic gene regulatory networks spanning roughly 18,000 genes and 300,000 regulatory edges, allowing us to isolate signaling dependencies that appear specifically during the resistance transition rather than in the stable pre- or post-treatment states. Candidate targets were cross-validated against genome-scale CRISPR dependency data, protein interaction networks, and pathway enrichment to eliminate transition-associated genes with no functional consequence. For the highest-confidence targets, we assessed ligandability using experimental and AlphaFold-predicted structures, then screened a virtual library exceeding one million drug-like compounds through a graph neural network prioritization step followed by docking, molecular dynamics, and binding free-energy calculation. Hits were filtered against blood-brain barrier permeability, CNS-MPO scoring, synthetic accessibility, and predicted toxicity to retain compounds with realistic pharmacological viability.