Neighborhood nonnegative matrix factorization identifies patterns and spatially-variable genes in large-scale spatial transcriptomics data

TL;DR

This paper introduces a new method called neighborhood NMF to analyze spatial patterns in large-scale gene expression data from tissues.

Contribution

The novel contribution is the development of NNMF, a scalable method for identifying overlapping multicellular gene programs in spatial transcriptomics.

Findings

NNMF identifies functionally coherent neighborhoods in heterogeneous cell data.

It scales to millions of cells and thousands of genes, outperforming hard clustering methods.

NNMF reveals immunologically relevant gene signatures in colorectal cancer data.

Abstract

Methods for identifying complex multicellular spatial neighborhoods do not scale to existing spatial transcriptomics data, and often divide tissues into distinct neighborhoods with hard borders. We develop neighborhood NMF (NNMF) that identifies functionally coherent neighborhoods among heterogeneous cells. NNMF scales to thousands of genes and millions of cells, and produces signatures representing overlapping spatially-organized multicellular gene programs, allowing more biologically-complex interpretations than hard clustering methods. In benchmark spatial transcriptomics data with expert labels, versus related methods, NNMF shows excellent performance even on hard clustering tasks. On MERFISH human colorectal cancer data, NNMF identifies immunologically relevant signatures in millions of cells.

The online version contains supplementary material available at 10.1186/s13059-025-03846-6.

Figures