Reproduce “An immune world model for multiscale forecasting and therapeutic hypothesis generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Taoyong Cui, Xi Wang, Zonghang Li, Jinchao Ding, Lingsen You, Yuzhi Xu, Wanghan Xu, Fang Wu and 6 more
An intervention-conditioned immune world model links cellular programs, tissue ecosystems, and patient-level response states to forecast how cytokines, drugs, genetic perturbations, and combinations alter immune biology. Across held-out perturbations and cross-scale tasks, it improves several forecasting benchmarks over established baselines and nominates IL-36γ with SIRPα inhibition as a complementary therapeutic hypothesis, while its self-correction audit rejects all screened cytokine pairs as insufficiently robust.
Key insights
- Separating state from action helps the model capture intervention-dependent effects rather than treating immune response as a fixed property of a cell or patient context.
- Multiscale information contributes progressively: combining cellular and tissue features raises immune-ecosystem classification macro-F1 from 0.52 and 0.60 individually to 0.64, while full-cell state improves anti-PD-1 response AUC to 0.92 in the reported cohort.
- The therapeutic audit favored complementary mechanisms—an innate CD47/SIRPα checkpoint axis and an adaptive/interferon-associated IL-36γ axis—but found no cytokine pair that met all robustness and dominance criteria.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.