SD2: spatially resolved transcriptomics deconvolution through integration of dropout and spatial information

• DOI: 10.1093/bioinformatics/btac605
• Publication Date: 2022-09-05T17:58:46Z
• Authors: Haoyang Li; Hanmin Li; Juexiao Zhou; Xin Gao
• Keywords: Dropout (neural networks); Benchmark (surveying)

Abstract Motivation Unveiling the heterogeneity in tissues is crucial to explore cell–cell interactions and cellular targets of human diseases. Spatial transcriptomics (ST) supplies spatial gene expression profile which has revolutionized our biological understanding, but variations cell-type proportions each spot with dozens cells would confound downstream analysis. Therefore, deconvolution ST been an indispensable step a technical challenge toward higher-resolution panorama tissues. Results Here, we propose novel method called SD2 integrating information data embracing important characteristic, dropout, traditionally considered as obstruction single-cell RNA sequencing (scRNA-seq) First, extract dropout-based genes informative features from scRNA-seq by fitting Michaelis–Menten function. After synthesizing pseudo-ST spots randomly composing data, auto-encoder applied discover low-dimensional non-linear representation real- spots. Next, create graph containing embedded profiles nodes, edges determined transcriptional similarity relationship. Given graph, convolutional neural network used predict compositions for real-ST We benchmark performance on simulated seqFISH+ dataset different resolutions measurements show superior compared state-of-the-art methods. further validated three real-world datasets technologies demonstrates capability localize composition accurately quantitative evidence. Finally, ablation study conducted verify contribution modules proposed SD2. Availability implementation The freely available github (https://github.com/leihouyeung/SD2) Zenodo (https://doi.org/10.5281/zenodo.7024684). Supplementary are at Bioinformatics online.