The acceleration of automated scientific discovery has been fundamentally bottlenecked by the epistemic gap between the semantic reasoning of large language models (LLMs) and the complex, non-linear reality of mammalian biology. While recent multi-agent frameworks have achieved autonomous hypothesis generation and in vitro experimental analysis, they frequently lack the rigorous statistical constraints required for multi-scale clinical translation. Furthermore, while algorithmic clinical digital twins successfully forecast biological states, they often rely on opaque latent spaces, sacrificing mechanistic interpretability for predictive accuracy. Here, we introduce the Multi-Scale Autonomous Discovery Engine (Octopus), a neuro-symbolic framework that unites a fully localised, privacy-preserving multi-agent swarm with regularised predictive algorithmic environments. Rather than stopping at isolated cellular assays, the system autonomously prioritises therapeutic hypotheses against in vitro CRISPR dependency data (CCLE), traces feature attribution cascades using XGBoost SHAP vectors, and orthogonally translates emergent vulnerabilities in silico to predict in vivo mammalian tumour trajectory (PDX) and human overall survival (Marisa). In a fully unsupervised sweep of colorectal cancer transcriptomes, the pipeline autonomously prioritised Insulin-like Growth Factor 2 (IGF2) as a significant candidate vulnerability to 5-Fluorouracil resistance.
The acceleration of automated scientific discovery has been fundamentally bottlenecked by the epistemic gap between the semantic reasoning of large language models (LLMs) and the complex, non-linear reality of mammalian biology. While recent multi-agent frameworks have achieved autonomous hypothesis generation and in vitro experimental analysis, they frequently lack the rigorous statistical constraints required for multi-scale clinical translation. Furthermore, while algorithmic clinical digital twins successfully forecast biological states, they often rely on opaque latent spaces, sacrificing mechanistic interpretability for predictive accuracy. Here, we introduce the Multi-Scale Autonomous Discovery Engine (Octopus), a neuro-symbolic framework that unites a fully localised, privacy-preserving multi-agent swarm with regularised predictive algorithmic environments. Rather than stopping at isolated cellular assays, the system autonomously prioritises therapeutic hypotheses against in vitro CRISPR dependency data (CCLE), traces feature attribution cascades using XGBoost SHAP vectors, and orthogonally translates emergent vulnerabilities in silico to predict in vivo mammalian tumour trajectory (PDX) and human overall survival (Marisa). In a fully unsupervised sweep of colorectal cancer transcriptomes, the pipeline autonomously prioritised Insulin-like Growth Factor 2 (IGF2) as a significant candidate vulnerability to 5-Fluorouracil resistance. The discovery maintained significance after rigorous Benjamini-Hochberg false discovery rate correction (q = 0.0292, Log-Rank p = 0.0007) and successfully predicted significant in vivo tumour volume shrinkage in an independent mouse cohort (Mixed-Effects LMM p = 0.0373). By bridging agentic hypothesis generation with statistically bounded clinical survival, this framework establishes a verifiable, local paradigm for the automated computational prioritisation of biomedical discoveries.