CM4AI-AnVIL - Cell Map Prediction DREAM Challenge
Scientific Background
The spatial organization of proteins within cells underpins nearly all cellular functions. Accurate models of protein localization would bridge the gap between molecular- and cellular-scale biology, enabling a more unified understanding of how cells operate. However, current imaging techniques, such as primary immunofluorescence, can visualize only tens of protein species at a time, whereas a single human cell contains thousands. AI-driven prediction methods can close this gap, enabling comprehensive cell maps at a scale existing imaging techniques alone cannot achieve. The goal of the CM4AI/AnVIL Cell Map Prediction DREAM Challenge is to bridge this gap. Participants will be presented with reference microscopy images and asked to predict additional channels revealing where specified proteins would localize within those cells. This Challenge brings together the massive amounts of data generated by the Bridge2AI project and the NHGRI's AnVIL cloud genomics platform. This challenge provides a unique opportunity to develop more flexible and generalizable generative image models that are capable of simulating the subcellular protein localization of a given protein based on its sequence, paving the way for a new era of in silico microscopy experiments.
CM4AI/AnVIL Cell Map Prediction DREAM Challenge
This challenge will advance the development of generative AI models capable of predicting protein subcellular localization from microscopy images. The primary benefit will be to establish new generative image models in biology that can simulate the subcellular distribution of a protein, while encouraging creative solutions for integrating multi-modal molecular data to enable spatially precise protein image prediction. Participating teams will be asked to generate IF microscopy images of target proteins given their amino acid sequences and reference channel images. See References [2, 3, 5-7] for relevant background publications.
A starter kit of training data will be available via the NHGRI AnVIL platform. This will include spatial data from the Human Protein Atlas [4], along with public CM4AI IF datasets covering 50,000 cells, each stained for one of 500 proteins [1]. Submitted models will be evaluated on an aggregated panel of performance metrics.
Cell Maps For AI
The Bridge2AI Cell Maps for AI (CM4AI) data generation project seeks to map the spatiotemporal architecture of human cells and use these maps toward the grand challenge of interpretable genotype-phenotype learning, essentially building "virtual cell" models. The CM4AI project has generated new flagship datasets for functional genomics using SEC-MS, AP-MS, and immunofluorescence (IF) data to develop cell maps for multiple cell lines [1].
AnVIL
The NHGRI AnVIL (Genomic Data Science Analysis, Visualization, and Informatics Lab-space) is a cloud-based platform designed for sharing, managing, and analyzing large-scale genomic datasets. The AnVIL is the NHGRI's collaborative computing platform in the cloud that hosts genomic datasets such as 1000 Genomes, GREGoR, eMERGE, and GTEx. It enables Cloud-Based Models, Scalable Computing, Data Management with Security that meets federal standards. Using the reproducible computing technology of the AnVIL platform, participants will submit models that can be applied to new datasets.
Challenge Incentives
A set of top-performing models will be eligible for monetary awards totaling $50,000 and cloud computing credits totaling $70,000 provided by the NHGRI, and co-authorship in the overall challenge paper.
Flyer
Event Details
- Event page: https://www.synapse.org/Synapse:syn77349038/wiki/643745
- Agenda: The challenge opens on October 19, 2026 and closes on February 1, 2027.
- How to register: Pre-register on the challenge page to receive updates about the challenge.