Binary-SPA: a reference-free method for cell annotation in high-resolution spatial transcriptomics

Abstract

Abstract Accurate cell annotation is a primary challenge in spatial transcriptomics (ST). Current approaches primarily rely on label transfer from single-cell RNA sequencing (scRNA-seq) reference or marker-based clustering. While these methods are widely used, they have critical limitations. Label transfer approaches depend on the availability of a well-matched scRNA-seq reference. Marker-based annotation methods often suffer from accuracy and limited coverage. To address these challenges, we developed Binary-SPA (Binary Self-referenced Projection Annotation), a computational framework for cell-type annotation of high-resolution ST. Binary-SPA performs annotation in two stages. First, a binary classification step identifies high-confidence cells using predefined marker sets. These confidently annotated cells are then used as an internal reference for anchor-based label transfer in the second stage. Binary-SPA consistently outperforms conventional marker-based methods and matches or exceeds label-transfer methods across diverse spatial platforms, tissue types, and preservation protocols—without requiring matched reference data. Unlike label transfer methods that decline without same-tissue references, Binary-SPA eliminates external data dependencies entirely while maintaining high accuracy at high annotation coverage. Binary-SPA demonstrates robust performance in challenging specimens such as bone marrow biopsies, and validation against matched COMET protein expression data confirmed strong concordance between transcriptomic- and protein-based cell identities. Binary-SPA thus provides a robust, reference-free solution for ST annotation with broad applicability to research and clinical specimens.