Contextual Invertible World Models: A Neuro-Symbolic Agentic Framework for Colorectal Cancer Drug Response

The article introduces the Contextual Invertible World Model (CIWM), a Neuro-Symbolic Agentic Framework designed to enhance precision oncology by integrating a machine learning emulator with a Large Language Model for reasoning. Utilizing the Sanger GDSC dataset with 83 samples, the framework achieves a baseline predictive correlation of \( r=0.268 \) for complex transcriptomics and employs Inverse Reasoning to uncover significant biological insights, such as the dominance of mutant KRAS in 5-fluorouracil resistance and the unintended consequences of PIK3CA repair on chemoresistance. This approach offers a novel mechanism for understanding drug responses in colorectal cancer, potentially guiding more effective therapeutic strategies.