Document resource
Background Advances in artificial intelligence for healthcare demand rich, high-volume datasets. While large-scale data atlases, particularly of single-cell RNA sequencing profiles, have proven invaluable for biological discovery, a widely accessible spatial proteomics atlas remains elusive. To date, the challenges of acquiring clinically viable samples, integrating diverse sample types, and harmonizing disparate experimental methods in multiplexed imaging have hindered the development and widespread availability of such atlases. Such an atlas could provide novel insights into tissue biology, enhance clinical trials, and aid in the identification of new drug targets.Methods To meet this need, we introduce the Enable PanCancer Atlas, a single-cell spatial proteomics database consisting of over 100 million segmented single cells from 8500+ independent samples, each imaged with 60 protein features generated with the Akoya Phenocycler and co-registered with same-slide H&E stains. Each sample in the Atlas contains rich clinical metadata, including patient demographics, treatment history and longitudinal time points, and survival outcomes. These multimodal characteristics allow researchers to create diverse cohorts and make comparisons within the database alone.Results We demonstrate the Atlas’ potential with a patient stratification case study across a cohort of 80 hepatocellular carcinoma samples. To do this, we leveraged SPARC (Scoring Pathways and ARChitectures), a set of deep learning-derived signatures that quantify the contribution of tissue microenvironments to immunotherapy outcomes. 1 For each sample, we defined cell types and used our previously trained model to infer SPARC tumor microenvironments. We used these scores to predict the effectiveness of checkpoint inhibition for each patient, and stratified patients into likely responder and nonresponder cohorts. This work demonstrates the potential of these calculated resistance signatures to aid clinical patient stratification and highlights the Atlas database’s utility for model validation in deep learning algorithms.Conclusions In conclusion, the Enable PanCancer Atlas is an unprecedented resource for spatial proteomics research, artificial intelligence model development, and personalized medicine in the omics era. Its richly featured database of diverse cancer types and stages allows for powerful discovery and validation analyses. We demonstrate its value for validating the extensibility of model signatures in the context of immunotherapy treatment in hepatocellular carcinoma. This work highlights the Atlas’ utility for biological discovery, to reinforce clinical trial results, and as a tool to create and validate deep learning models in spatial biology.Reference Wu Z, et al. Spatial multi-omics and deep learning reveal fingerprints of immunotherapy response and resistance in hepatocellular carcinoma. bioRxiv. 2025.Ethics Approval All data generation was performed on archival deidentified tissue.