spora:用于空间蛋白质组学的统一多模态数据集与平台
A multimodal dataset for spatial proteomics - ΑΙhub
EPFL 团队发布空间蛋白质组学数据集与平台 spora,收录来自 5,477 名患者的 13,546 张多重成像图像,覆盖 14 种组织来源和五种成像技术。该资源保留原始研究与标注的关联,并提供数据访问、分析、AI 模型开发及跨研究评估工具;相关研究发表于 NeurIPS 2026。
正文
A multimodal dataset for spatial proteomics
Imaging mass cytometry of a lung tumor core from a tissue microarray. 2026 EPFL/AIMM – CC-BY-SA 4.0.
By Tanya Petersen
Researchers at EPFL have created the world’s largest open-access collection of spatial proteomics tissue images, bringing together samples from more than 5,000 patients. Called spora, this new resource aims to accelerate AI research into how tissues function and how diseases develop.
How cells are arranged within tissue can reveal as much about disease as the cells themselves. When doctors examine a cancer tumor, for example, they don’t just look at which cells are present, they also want to know how those cells are arranged and how they interact with each other. That’s because the way cancer cells interact with neighboring immune cells can offer important clues about how a disease develops or responds to treatment.
Scientists can study these relationships using spatial proteomics, a technology which captures this organization by measuring dozens of proteins while preserving their locations. Because proteins help control what cells do, these images offer a valuable picture of how healthy and diseased tissues work.
But the resulting datasets are difficult to combine. Research laboratories around the world use different equipment, examine different proteins and store their results in different ways. Combining this information is complicated and time-consuming, making it difficult to use large amounts of data in both conventional studies, and now to train artificial intelligence.
To overcome this challenge, a team led by Charlotte Bunne’s Artificial Intelligence in Molecular Medicine group at EPFL, in both the School of Life and the School of Computer and Communication Sciences, has developed spora, a spatial proteomics reference atlas – the world’s largest, freely available collection of spatial proteomics tissue images and medical data, designed to make research easier.
Two years of work
Released in collaboration with the European Molecular Biology Laboratory’s European Bioinformatics Institute (EMBL-EBI), the collection includes 13,546 multiplex images from 5,477 patients, spanning 14 tissue origins and five imaging technologies. Where available, it also includes paired histopathology images, connecting molecular measurements with the tissue images used in routine pathology.
Creating spora took two years of work by scientists and biomedical experts from EPFL, University Hospital Zurich, the University of Zurich, Geneva University Hospitals and the University of Geneva. The researchers had to organize thousands of images, standardize protein annotations, match medical information and identify individual cells so that data from different studies could be compared.
“Gathering and harmonizing data across cohorts takes enormous time, creating a major hurdle for both large-scale AI development and biological discovery,” says Benedikt von Querfurth, a doctoral student at EPFL and first author of the study. “We want to allow researchers to start with ready to use data so that they can focus on the models and biological questions.”
Toward better predictions for the patient
The collection retains links to the original studies and their annotations. Differences in the proteins measured, imaging conditions, and available clinical information remain visible, helping researchers decide which datasets are appropriate for their questions.
In addition to the ‘atlas’ itself, spora provides researchers with tools to access and analyze the data, develop AI models and evaluate how well they perform across different studies. “This is important because a model can perform well within one study and still struggle when the patients, laboratory, or measurement technology change,” says Lukas Klein, a PostDoc at EPFL and the University Hospital Zurich. “Shared datasets and evaluation protocols let us examine those differences systematically. That is essential if we want AI models to capture biology that extends beyond a single cohort.”
Although spora currently focuses mainly on cancer research, the team hopes to expand it to include more tissues, diseases and imaging technologies. “Our goal is to turn what we learn across thousands of tissue samples into insights that can help individual patients,” says Bunne. “Working closely with clinicians, we aim to translate these insights into better predictions of how a patient’s disease will progress and which treatments are most likely to benefit them.”
Reference
spora: A Unified Multimodal Dataset and Platform for Spatial Proteomics, Benedikt von Querfurth, Johann Wenckstern, Eeshaan Jain, Lukas Klein, Phil Cheng, Yexiang Cheng, Cedric Vincent-Cuaz, Martina Haberecker, Pascal Frossard, Andreas Wicki, Olivier Michielin, Charlotte Bunne, NeurIPS 2026.
Controllable AI and the power of local knowledge: an interview with Carmen Amo Alonso
Ella Scallan
17 Nov 2026
How can we build AI systems that do what we want them to do? Control theory can help.
Data-driven re-discovery of 2D materials
Linköping University
16 Nov 2026
With the help of AI, researchers have found previously overlooked occurrences of the 2D materials MXenes in older databases.
Tech-savvy STEM students prone to letting AI do their thinking for them
Oregon State University
13 Nov 2026
How can students break their dependence on AI tools?
An invisible word of the day: watermarking assignment prompts to detect copy-pasted chatbot answers
Aizierjiang Aiersilan
12 Nov 2026
SteganoPrompt lets educators embed an invisible but openly announced instruction in an assignment prompt.
AI100 longitudinal study examines how AI has entered the mainstream
Lucy Smith
10 Nov 2026
Published every five years, the report is compiled by experts in the field.
AI pioneers
Thinking big – from constraint programming to the Holy Grail: an interview with Gene Freuder
Ella Scallan
10 Nov 2026
In our latest instalment of AI Pioneers, we speak to Gene Freuder on the field of constraint programming, neurosymbolic and generative AI approaches in AI, and the bigger picture of computer science.
How attackers persuade AI agents to break the rules
EPFL
09 Nov 2026
As AI assistants evolve into AI agents, a study has found that the biggest safety risks in their use may not come from single malicious prompts, but from carefully orchestrated conversations.
Interview with Minji Park: humans in the loop to keep language models up to date
Lucy Smith
06 Nov 2026
Human-AI co-inference through a shared editable state.
Association for the Understanding of Artificial Intelligence
AI 热点