{"title":"Beyond Co-Localization: DeepCoEx Learns Exclusion-Aware Spatial Representations in Mass Spectrometry Imaging via Dual-Contrastive Learning","authors":"Jingjing Xu,Xin Diao,Siyang Liu,Thomas Ka-Yam Lam,Yusong Zhang,Shuang Gao,Lingli Deng,Weiji Chen,Lei Guo,Zongwei Cai,Jiyang Dong","doi":"10.1021/acs.analchem.6c02338","DOIUrl":null,"url":null,"abstract":"Abstract Mass spectrometry imaging (MSI) provides spatially resolved molecular profiles, yet the existing computational workflows are heavily biased toward co-localization and remain limited in characterizing exclusion-like spatial relationships, especially when reliable negative labels are scarce. To address this challenge, we present DeepCoEx (Deep Co-localization and Exclusion), a self-supervised framework that learns exclusion-aware spatial representations through dual-contrastive learning for spatial metabolomics analysis. Instead of relying on large collections of annotated negative ion pairs, DeepCoEx embeds positive spatial consistency and exclusion-aware separation into a unified representation space. This design enables the model to preserve spatially coherent ion patterns while improving the separability of spatially complementary distributions. Compared with conventional similarity metrics and dimensionality reduction baselines, DeepCoEx provides more consistent and interpretable retrieval of candidate negative spatial associations. In representative control and PM2.5-exposed rat brains, the framework reveals localized spatial redistribution patterns and brainstem-related complementary ion associations that are not readily captured by conventional correlation-based analysis. In barley seeds, the learned embeddings help resolve chemically related ion species, including protonated and sodiated adducts of phosphatidylcholine species and glycosylated hordatine derivatives. Furthermore, DeepCoEx supports subregion-specific network construction, enabling exploratory analysis of localized molecular association patterns under different tissue or treatment contexts. By establishing an exclusion-aware representation learning strategy for MSI, DeepCoEx expands spatial metabolomics beyond co-localization and provides a practical workflow for analyzing complementary molecular organization in complex tissues.","PeriodicalId":27,"journal":{"name":"Analytical Chemistry","volume":"24 1","pages":""},"PeriodicalIF":7.3000,"publicationDate":"2026-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Analytical Chemistry","FirstCategoryId":"92","ListUrlMain":"https://doi.org/10.1021/acs.analchem.6c02338","RegionNum":1,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"CHEMISTRY, ANALYTICAL","Score":null,"Total":0}
引用次数: 0
Abstract
Abstract Mass spectrometry imaging (MSI) provides spatially resolved molecular profiles, yet the existing computational workflows are heavily biased toward co-localization and remain limited in characterizing exclusion-like spatial relationships, especially when reliable negative labels are scarce. To address this challenge, we present DeepCoEx (Deep Co-localization and Exclusion), a self-supervised framework that learns exclusion-aware spatial representations through dual-contrastive learning for spatial metabolomics analysis. Instead of relying on large collections of annotated negative ion pairs, DeepCoEx embeds positive spatial consistency and exclusion-aware separation into a unified representation space. This design enables the model to preserve spatially coherent ion patterns while improving the separability of spatially complementary distributions. Compared with conventional similarity metrics and dimensionality reduction baselines, DeepCoEx provides more consistent and interpretable retrieval of candidate negative spatial associations. In representative control and PM2.5-exposed rat brains, the framework reveals localized spatial redistribution patterns and brainstem-related complementary ion associations that are not readily captured by conventional correlation-based analysis. In barley seeds, the learned embeddings help resolve chemically related ion species, including protonated and sodiated adducts of phosphatidylcholine species and glycosylated hordatine derivatives. Furthermore, DeepCoEx supports subregion-specific network construction, enabling exploratory analysis of localized molecular association patterns under different tissue or treatment contexts. By establishing an exclusion-aware representation learning strategy for MSI, DeepCoEx expands spatial metabolomics beyond co-localization and provides a practical workflow for analyzing complementary molecular organization in complex tissues.
期刊介绍:
Analytical Chemistry, a peer-reviewed research journal, focuses on disseminating new and original knowledge across all branches of analytical chemistry. Fundamental articles may explore general principles of chemical measurement science and need not directly address existing or potential analytical methodology. They can be entirely theoretical or report experimental results. Contributions may cover various phases of analytical operations, including sampling, bioanalysis, electrochemistry, mass spectrometry, microscale and nanoscale systems, environmental analysis, separations, spectroscopy, chemical reactions and selectivity, instrumentation, imaging, surface analysis, and data processing. Papers discussing known analytical methods should present a significant, original application of the method, a notable improvement, or results on an important analyte.