NEW YORK – Backed by $36.5 million in funding from the Knut and Alice Wallenberg Foundation, researchers at Sweden's SciLifeLab and the KTH Royal Institute of Technology have launched an effort to longitudinally model cellular components and processes.
The project, called Alpha Cell, will build on proteomic data generated by KTH's Human Protein Atlas (HPA) and will combine that data with AI analyses to generate three-dimensional models of cellular structures and track their behavior over time and in response to different perturbations, said Mathias Uhlén, professor of microbiology at KTH and leader of the HPA and the Alpha Cell initiative.
"What we want to do is try to harness the AI revolution to predict and explain living systems," Uhlén said.
The initial stage of the project will be what Uhlén called an annotation exercise. "The focus in the beginning will very much be to try to define the buildings blocks of human cells — what are the housekeeping proteins, what are the tissue-specific proteins, what is the functional proteome, and so on."
From there, Uhlén said, the researchers hope to define the protein networks and interactions present in different cellular compartments and across different time points.
The project will leverage the proteomic data that has been generated through the HPA. Launched in 2000, the HPA has worked to map human proteins in a variety of cells and tissues, as well as blood, primarily using antibody-based technologies but also mass spectrometry and transcriptomics.
The most recent version of the atlas was launched in October at the annual meeting of the Human Proteome Organization. The atlas features more than 10 million manually annotated bioimages as well as single-cell gene expression profiles across 81 cell types spanning 31 human tissues and protein localization data for 1,021 proteins, including localization data specific to tissues, single-cell types, particular cell states, and subcellular structures. The HPA's subcellular resource currently includes expression and spatiotemporal data for 13,534 proteins, with subcellular distribution data for each of those proteins generated in up to three different cell lines and classification of the proteins as part of one or more of 49 organelles or subcellular structures.
Beyond the HPA data, the project also hopes to incorporate proteomic data from other initiatives around the world, Uhlén said, citing as an example the Chinese π-HuB Consortium, or Proteomics Navigator of the Human Body, which similarly aims to map the human proteome at levels ranging from individual cell types to tissues and organs.
Jan Ellenberg, director of SciLifeLab and Uhlén's collaborator on the Alpha Cell project, said that he and his colleagues plan to use imaging technologies, including advanced fluorescence microscopy, cryogenic electron microscopy, and spatial omics imaging, to "generate high-resolution, dynamic views of cellular macromolecules and their interactions." He added that methods including super-resolution microscopy and light-sheet imaging would allow the researchers to "visualize molecular events over time in live cells."
The project aims to "systematically map organelles, protein complexes, chromatin architecture, membrane topology, and other subcellular structures with unprecedented spatial and temporal resolution," Ellenberg said. By integrating that data with AI-driven annotation tools, he and his team will be able to automate highly detailed identification and categorization of cellular components.
Like Uhlén, Ellenberg highlighted the HPA's datasets as a key resource, noting they provide "a unique foundation for training AI models to enhance protein annotation, improve structure-function predictions, and refine cellular mapping."
Generally speaking, advances in AI analyses have allowed researchers to automate what were once labor-intensive tasks with limited scalability, Ellenberg said, noting that AI-based bioimage analysis pipelines can, for instance, segment and classify cellular compartments, fit structural models and infer conformational changes, track dynamic molecular interactions in real time, and infer protein activity from imaging data.
He added that the SciLifeLab's AI infrastructure "is evolving rapidly" and suggested that future developments will include real-time AI-guided imaging, where models dynamically adjust microscope settings based on live data analysis and automated hypothesis generation where AI suggests new experiments based on its analyses.
The $36.5 million in funding from the Wallenberg Foundation adds to what Uhlén said was a substantial increase in government funding for the SciLifeLab included in the Swedish government budget that was released at the beginning of the year. That budget added roughly $10 million in annual funding to the existing funding for the SciLifeLab, which Uhlén said has averaged around $60 million per year. He said that at the beginning of February, the Stockholm-based organization decided to use those funds in part to construct a third building to house its expanding operations. He said the lab plans to add around 200 to 300 employees by the end of the year, which will bring its total staff to between 1,700 and 1,800 people.
Uhlén added that with the new Wallenberg funding, the HPA project is funded through 2030.