AtlasMap: enabling low-cost, map-style exploration of million-cell single-cell atlases

Abstract

Interactive visualization is critical for interpreting single-cell atlases, yet existing web-based tools struggle to handle the growing scale of multimillion-cell datasets, often constrained by browser memory limits and rendering latency. Here, we present AtlasMap, a scalable visualization framework that overcomes these bottlenecks through a multi-resolution, tile-based architecture. Unlike conventional systems that transfer cell-level data to the client, AtlasMap employs an offline preprocessing module to generate quadtree-based spatial aggregations stored in compressed Zarr v3 formats. A high-performance Go backend dynamically renders these summaries as PNG tiles, allowing a lightweight frontend to support fluid pan-and-zoom exploration similar to digital geographic maps. Systematic benchmarking against leading tools (including cellxgene and UCSC Cell Browser) demonstrates that AtlasMap decouples visualization performance from dataset size. While point-based approaches incurred prohibitive client-side memory costs or failed entirely at the 11-million-cell scale, AtlasMap maintained sub-second startup latency and a negligible browser footprint (<5 MB). By shifting computational pressure from the browser to an optimized server-side pipeline, AtlasMap enables accessible, high-fidelity exploration of ultra-scale single-cell datasets on standard hardware.