This article introduces TrimNN, a novel graph-based deep learning framework designed to characterize cellular community (CC) motifs within complex tissues using spatial transcriptomics and proteomics data. Unlike traditional top-down clustering approaches, TrimNN employs a bottom-up strategy to identify conserved spatial patterns of cell interactions, offering improved interpretability and generalizability. The research demonstrates TrimNN's effectiveness in revealing phenotype-associated CC motifs in studies of colorectal cancer and Alzheimer's disease, showing its potential for discovering biological and pathological mechanisms and identifying prognostic biomarkers. Furthermore, TrimNN proves to be accurate, scalable, and robust against various noise levels, outperforming existing methods in identifying and quantifying these intricate multicellular organizational patterns.

References:

• Yu Y, Wang S, Li J, et al. TrimNN: Characterizing cellular community motifs for studying multicellular topological organization in complex tissues[J]. Research Square, 2025: rs. 3. rs-5584635.