ID #826 Physics-Informed Neural Network–Guided Identification of Synthetic Resistance Collapse Points and RNA–Small Molecule Chimera Therapeutics to Overcome Adaptive Therapy Escape in Glioblastoma Multiforme

Abstract

Glioblastoma multiforme remains uniformly lethal largely due to rapid therapeutic adaptation, where tumor cell populations dynamically rewire signaling, metabolism, and transcriptional states to evade cytotoxic and targeted therapies. Adaptive therapy strategies attempt to exploit evolutionary tradeoffs, yet escape remains inevitable because resistance trajectories are neither static nor linear. This work presents an integrated computational and therapeutic framework that uses physics informed neural networks to identify synthetic resistance collapse points and rationally design RNA–small molecule chimera therapeutics to preempt adaptive escape in glioblastoma.

We develop a multiscale physics informed neural network that explicitly encodes tumor growth kinetics, drug diffusion, metabolic flux constraints, and regulatory network dynamics governing stemness, DNA damage response, and epigenetic plasticity. Unlike purely data driven models, the network is constrained by mechanistic differential equations describing cell state transitions, fitness landscapes, and therapy induced selective pressures. Trained on longitudinal transcriptomic, single cell RNA sequencing, and pharmacologic response data from glioblastoma models, the framework reconstructs hidden resistance manifolds and predicts critical bifurcation points at which compensatory pathways become mutually dependent and fragile.

From these analyses, we define synthetic resistance collapse points as parameter regimes where simultaneous perturbation of RNA regulatory nodes and enzymatic effectors produces irreversible loss of adaptive capacity. To exploit these points therapeutically, we propose a new class of RNA–small molecule chimera constructs that couple sequence specific RNA targeting, including long noncoding RNAs and resistance associated splice variants, with small molecule warheads directed against metabolic or signaling enzymes. The physics informed model guides chimera selection by optimizing timing, dosage, and target pairing to ensure intervention occurs precisely when evolutionary escape routes are maximally constrained.

In silico perturbation experiments demonstrate that PINN guided intervention collapses resistant subclones, suppresses phenotypic plasticity, and prevents rebound growth across heterogeneous tumor populations.

1. Gene Expression Omnibus. (n.d.). GSE77307: Transcriptomic analysis of U87-MG glioblastoma cells under hypoxic conditions. National Center for Biotechnology Information. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE77307 | Yang, Chendong et al. “Analysis of hypoxia-induced metabolic reprogramming.” Methods in enzymology vol. 542 (2014): 425-55. doi:10.1016/B978-0-12-416618-9.00022-4 Gonzalez, Frank J et al. “The role of hypoxia-inducible factors in metabolic diseases.” Nature reviews. Endocrinology vol. 15,1 (2018): 21-32. doi:10.1038/s41574-018-0096-z Infantino, Vittoria et al. “Cancer Cell Metabolism in Hypoxia: Role of HIF-1 as Key Regulator and Therapeutic Target.” International journal of molecular sciencesvol. 22,11 5703. 27 May. 2021, doi:10.3390/ijms22115703 Belisario, Dimas Carolina et al. “Hypoxia Dictates Metabolic Rewiring of Tumors: Implications for Chemoresistance.” Cells vol. 9,12 2598. 4 Dec. 2020, doi:10.3390/cells9122598 Singhal, Rashi, and Yatrik M Shah. “Oxygen battle in the gut: Hypoxia and hypoxia-inducible factors in metabolic and inflammatory responses in the intestine.” The Journal of biological chemistry vol. 295,30 (2020): 10493-10505. doi:10.1074/jbc.REV120.011188