{
  "id": 10450685,
  "title": "GEMOT: Towards Mechanistic World Models for Biology",
  "url": "https://urgent.news/2026/09/28/gemot-towards-mechanistic-world-models-for-biology",
  "topic": "science",
  "section": "Science",
  "published": "2026-09-28T00:00:00.000Z",
  "source": {
    "name": "bioRxiv",
    "slug": "biorxiv",
    "url": "https://www.biorxiv.org/content/10.64898/2026.09.26.754619v1?rss=1"
  },
  "original_language": "en",
  "account": "The pursuit of scientific discovery requires understanding the mechanisms that underlie observations and enable predictions beyond the initial data. In contrast to large language models, which implicitly encode knowledge, mechanistic world models structure this knowledge into explicit, modular, and reusable elements. These models can have their predictions evaluated against data, providing a clear scoring mechanism.\n\nIn the field of biology, measurements can be limited and noisy. Consequently, mechanistic world modelling must simultaneously uncover hidden states and the governing equations that govern them. However, the prior knowledge that could guide this search is often unstructured. To address this challenge, the authors introduce gemot, an agentic framework for mechanistic world modelling.\n\nThe gemot framework is evaluated using 18 published biological problems, covering molecular biology, epidemiology, and immune-cell differentiation. These problems involve up to 1,755 training measurements, 65 observables, 170 experimental conditions, and three data modalities for each problem. What sets gemot apart is its auditable nature. A semantic layer records each hypothesis, while a Model Context Protocol layer separates hypothesis formulation from numerical evaluation, ensuring that hypothesis scoring cannot be fabricated.\n\nAcross all evaluated problems, models constructed autonomously by gemot either match or surpass reference models in terms of fit and parsimony. Furthermore, these models demonstrate the ability to generalize when held-out data is available. Perhaps most importantly, gemot has been shown to formulate novel, biologically plausible mechanistic hypotheses, expanding our understanding of biological systems.\n\nThe implications of this work extend beyond just biology. By providing a method to decode interventional biological data into competing mechanistic hypotheses, gemot can guide the design of experiments that can differentiate between these hypotheses. This approach may serve as a template for applying mechanistic world modelling techniques to other domains, paving the way for more comprehensive and interpretable scientific discoveries.",
  "summary": "Scientific discovery seeks mechanisms that explain observations and predict beyond the measurements that produced them. Whereas large language models (LLMs) encode knowledge implicitly, mechanistic world models organise it as a parsimonious set of explicit, modular, reusable mechanisms whose predictions can be scored against data. In biology, where measurements are sparse and noisy, mechanistic…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}