Deep Subspace Reconstruction-Driven Adaptive Canonical Correlation Analysis Model for Identifying Imaging Genetics...

This study introduces a two-stage correlated analysis model to overcome clinical information oversights in raw data manipulation for canonical correlation analysis extensions. The novel algorithm integrates whole slide images, gene expression data, and pathway scoring data to unveil high-order correlations linked to sarcoma recurrence. This study employs data from 259 sarcoma samples (TCGA and UCSC databases). A two-stage analysis was introduced, using a deep self-reconstruction model for multi-modal data integration, followed by a hypergraph-based adaptive sparse multi-view canonical correlation analysis to explore higher-order correlations among modal features. This study validates DSR-AdaSMCCA's effectiveness with error variance and correlation coefficient analyses, demonstrating faster convergence and higher coefficients, confirming the success of the deep subspace reconstruction strategy. Bioinformatics analysis confirms the algorithm's ability to discover the genes enriched in the sarcoma recurrence-related diseases and uncovered potential biological mechanisms in predicting sarcoma recurrence through the association between WSI features and genetic characteristics. The proposed model in this study successfully integrates imaging genetics data, accurately identifies key features associated with local recurrence of sarcoma, and reveals pathways closely linked to immune response and inflammation through enrichment analysis. The study deepens the understanding of sarcoma recurrence, providing valuable insights for personalized treatment strategies and unraveling the intricate networks influencing tumor relapse. The complete source code is openly available at https://github.com/babykai12345/HB-AdaSMCCA .