Helmholtz Munich and Ludwig Maximilians University have unveiled MouseMapper, a deep-learning system that reconstructs an entire mouse body at cellular resolution. Published in Nature on May 22, 2026, the framework already overturned a long-held assumption about obesity by exposing previously invisible damage to the facial sensory nerves of high-fat-diet mice. The same molecular fingerprint also showed up in human trigeminal tissue, hinting that a metabolic disease many considered purely visceral may quietly wear down peripheral neurons in people, too.
What Happened
The team led by Prof. Ali Ertürk, director of the Institute for Biological Intelligence at Helmholtz Munich, built MouseMapper on a foundation-model backbone trained to segment 31 organs and tissues across whole-body imaging volumes. After applying tissue-clearing chemistry to make mice optically transparent and tagging neurons and immune cells with fluorescent markers, the team captured each animal with light-sheet microscopy, generating datasets that span tens of millions of cellular structures per mouse. The AI then parsed those volumes automatically, producing 3D maps that no curated region-of-interest analysis could match.
The headline biomedical discovery emerged when the researchers fed mice a high-fat diet to mimic human obesity. MouseMapper flagged widespread inflammation across adipose tissue, muscle, liver and peripheral nerves, and singled out a striking pattern in the trigeminal nerve — the large facial nerve responsible for touch and certain motor functions. In obese animals the trigeminal showed major reductions in branching and nerve endings, a structural change the team later confirmed with behavioral tests in which obese mice were less responsive to sensory stimulation than their lean counterparts. The Nature paper is registered as DOI 10.1038/s41586-026-10535-2.
Why It Matters
Obesity is well known to raise the risk of type 2 diabetes, cardiovascular disease, stroke, neuropathy and several cancers, yet biologists have long been forced to study one organ at a time. MouseMapper's strength is that it does not ask scientists to pick where to look. By treating the whole mouse as the unit of analysis, the system uncovers cross-organ connections that traditional histology slides simply cannot capture. According to co-first author Ying Chen, the foundation-model design lets MouseMapper generalize far beyond the slices it was originally trained on, which is what allowed the trigeminal finding to surface without anyone explicitly hunting for it.
.jpg?width=1024)
The pipeline also demonstrates how foundation models — neural networks pretrained on broad data and then fine-tuned to new tasks — are migrating from language into biology. Earlier whole-mouse imaging projects relied on bespoke segmentation tools that struggled to keep up with the deluge of light-sheet data. By contrast, MouseMapper handles arbitrary tissue contexts in a single pass and outputs structured 3D annotations that downstream researchers can query, slice or feed into other models. That puts it in the same lineage as systems like Google DeepMind's protein and cell-state models, but pushed to the level of an intact organism rather than a single molecule.
Reaction
Inside the lab, senior scientist Dr. Doris Kaltenecker — first author of the study and an investigator at Helmholtz Munich's Institute for Diabetes and Cancer — framed the trigeminal observation as proof that integrated imaging changes what counts as a question worth asking. "We revealed previously unknown structural and molecular changes in the trigeminal ganglion and its facial branches, and the same molecular signature was conserved in human tissue. This kind of finding simply cannot emerge from studying one organ at a time," she said in the Helmholtz statement carried by ScienceDaily.
Outside the consortium, obesity and neurology specialists have taken the trigeminal result as a serious prompt to revisit the sensory side effects that patients sometimes describe but that rarely make it into clinical guidelines. Medical-imaging outlets including Medical Xpress highlighted that the same molecular signatures showed up in human trigeminal samples, raising the possibility that obesity-driven facial nerve remodeling has been underdiagnosed because no one had a tool to look for it across the whole body at once.
What's Next
The team has already deposited the whole-body datasets in a public portal so that other research groups can run their own analyses against the maps. According to the paper, the next priority is to extend MouseMapper to a wider catalog of diseases that span multiple organ systems — diabetes, cancer, neurodegenerative disorders and autoimmune conditions — and to push the framework toward inflammatory disease models that current single-organ workflows handle poorly. The European Research Council's CALVARIA grant and several German Excellence Strategy clusters are already funding follow-on projects.
Ertürk also sketched a longer arc that places MouseMapper inside the wider push toward biological digital twins. "Our long-term vision is to build truly realistic digital twins of mice in health and disease: cell-level atlases that we can query, perturb and screen in silico computationally," he told Helmholtz Munich. "That would let us pinpoint the earliest changes a disease causes, design interventions to prevent them, and accelerate the discovery of new treatments while reducing the number of physical experiments we need to run." In other words, the ambition is to turn each new mouse cohort into a permanent in silico reference, not a one-off experiment.
Closing Thoughts
MouseMapper is a useful corrective to the current AI-applications conversation, which often gravitates toward language-model demos and away from the harder work of grounding models in physical reality. By coupling foundation-model architectures with transparent-tissue imaging and behavioral validation, the Helmholtz team showed how a single AI tool can both confirm what physicians already suspected about obesity's systemic reach and surface entirely new biology — facial nerve damage — that targeted assays would have missed.
What makes the result feel durable is the public dataset release. Whole-body atlases that anyone can re-query lower the barrier for follow-up studies and let smaller labs ask whole-organism questions that previously required a microscopy facility and a custom segmentation pipeline. If the next wave of biological digital twins really does begin with frameworks like MouseMapper, the most important AI-applications headlines of the next year may come less from chatbot benchmarks than from quiet papers in which a foundation model points biologists toward an organ system they never thought to inspect.
한글 요약
독일 헬름홀츠 뮌헨과 루드비히 막시밀리안 대학 연구진이 2026년 5월 22일 Nature에 공개한 'MouseMapper'는 마우스 한 마리의 전신을 세포 단위로 자동 분석하는 파운데이션 모델 기반 AI다. 조직 투명화와 광시트 현미경으로 확보한 거대한 3D 영상 데이터를 다루며, 한 번의 추론으로 31개 장기와 신경·면역세포까지 분할해 낸다.
연구팀이 고지방 식이로 비만을 유도한 마우스를 MouseMapper로 스캔한 결과, 지방·근육·간뿐 아니라 안면 감각을 담당하는 삼차신경에서도 가지 수와 말단이 크게 줄어든 것이 확인됐다. 행동 실험에서도 비만 마우스의 감각 반응이 둔화됐고, 동일한 분자 시그니처가 사람 삼차신경 조직에서도 발견돼 비만이 신경계까지 광범위하게 영향을 미친다는 점이 처음으로 통합적으로 드러났다.
연구진은 전신 데이터셋을 공개해 다른 연구실도 활용할 수 있도록 했고, 당뇨·암·신경퇴행성 질환 등 다장기 질환으로 적용 범위를 넓힐 계획이다. 장기적으로는 마우스의 '디지털 트윈'을 구축해 신약 후보의 영향을 컴퓨터 상에서 미리 평가하고 동물실험을 줄이겠다는 목표를 함께 제시했다. 참고: ScienceDaily, Nature DOI.