By- Christina Elston
LOS ANGELES (May 8, 2026) — A team led by Cedars-Sinai Health Sciences University investigators has created a faster, cheaper way to determine the genes expressed in cancerous tumors. The AI-based tool, which they describe in Cell, could make personalized cancer treatment available to more patients.
The new tool, called Path2Space, predicts gene expression across the tumor area based on digital images of biopsy slides, which contain thin slices of tumor tissue that can be examined under a microscope.
Because tumors do not have the same composition and gene expression throughout, Path2Space predicts what is known as “spatial” gene expression, estimating it at many different points within the tumor. The process takes only minutes and costs significantly less than conventional spatial gene expression profiling, which typically takes several weeks and costs thousands of dollars.
“This tool makes two major contributions,” said Eytan Ruppin, MD, PhD, deputy director of the Translational Research Institute at Cedars-Sinai and senior author of the study. “It will enable us and others to study larger datasets and understand the spatial structure of tumors. But what really motivates me is that, if we can successfully validate the tool in clinical trials, it could improve cancer care for patients.”
Investigators “trained” Path2Space using data from a large group of patients with breast cancer, where the biopsy slides and spatial sequencing were both available. They then tested the tool on three additional patient datasets to validate its performance.
“For each sample, we looked at the actual, measured gene expression and compared it with our tool’s prediction,” said Eldad Shulman, PhD, co-first author of the study and a research fellow at the National Cancer Institute, who will soon join Ruppin’s lab as a research scientist. “For each sample, we predicted the spatial expression of almost 5,000 genes, and the predictions matched the measured expression well across all three patient groups.”
Path2Space is also designed to help scientists discover new biomarkers that could guide treatment decisions and identify patients at higher risk of poor outcomes.
“The tool looks at characteristics within the tumor, such as whether a gene is expressed in some areas of the tumor and not others,” said Emma Campagnolo, co-first author of the study and a research fellow in Ruppin’s lab. “We found specific spatial patterns of gene activity in tumors that predict how patients respond to treatment.”
Identifying spatial biomarkers is challenging, Shulman said, because the high cost of spatial profiling by traditional methods means very little of this data is available.
“Before we developed Path2Space, the largest cohort we could find to study the spatial organization of the tumor environment was about 30 patients,” Shulman said. “With this tool, we can study slides from thousands of patients. Path2Space is tapping into the potential of spatial biology in a way that has not been possible until now.”
Path2Space could be applied to other cancer types once it is trained on the correct data, and the lab is finalizing a study applying it to head and neck cancer, Campagnolo said. The team is also working to make the tool more precise. It currently looks at groups of 10 to 20 cells together, and the goal is to eventually be able to assess individual cells.
“With the help of clinical collaborators, we next want to bring Path2Space into clinical trials,” Ruppin said. “It represents an exciting development in a growing field and has to be tested carefully. But we are hopeful that it could make an impactful contribution to science and to patient care.”
Robert Figlin, MD, interim director of Cedars-Sinai Cancer, noted that translational research is a hallmark of the institution.
“The development of tools that apply leading-edge science to patient care is the best way to serve our patients—and to improve cancer care on a global scale,” Figlin said.
Additional Cedars-Sinai authors include Yuan Yuan, Karine Sargsyan, and Simon R.V. Knott.
Other authors include Roshan Lodha, Youngmin Chung, Amos Stemmer, Thomas Cantore, Beibei Ru, Tian-Gen Chang, Sumona Biswas, Saugato Rahman Dhruba, Sumeet Patiyal, Sushant Patkar, Andrew Wang, Ranjan K. Barman, Chuhan Wang, Rohit Paul, Sarath Chandra Kalisetty, Tom Hu, MacLean P. Nasrallah, Ellis Patrick, Jean Yang, Amy Plotkin, Padma Sheila Rajagopal, Stephen-John Sammut, Stanley Lipkowitz, Peng Jiang, Carlos Caldas, Kenneth Aldape, Joo Sang Lee, and Danh-Tai Hoang.
Funding: This research was supported by the Intramural Research Program of the NIH, NCI, and the Center for Cancer Research. The contributions of the NIH authors were made as part of their official duties, as NIH federal employees are in compliance with agency policy requirements and are considered works of the U.S. government. This research was also partially supported by a grant of the Korea-United States Collaborative Research Fund, funded by the Ministry of Science and ICT and the Ministry of Health and Welfare, Republic of Korea (grant number: RS-2024-00468417; Y.C. and J.S.L.), and by an Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (RS-2019-II190421, AI Graduate School Support Program, Sungkyunkwan University; Y.C. and J.S.L.). This work has utilized the computational resources of the NIH HPC Biowulf cluster.
Competing interests: E.D.S., E.M.C. and E.R. are listed as inventors on a provisional patent (application no. 63/703,060, United States, 2024) filed based on the methodology outlined in this study. E.R is (non-paid) member of the scientific advisory boards of Pangea Biomed (divested), GSK Oncology and the ProCan project. E.R is a founder of MedAware Ltd. The other authors declare no competing interests.
