Abstract 4193: Heterogeneous graph neural network meta-analysis of lung cancer and tuberculosis transcriptomic datasets reveals convergent host response networks

Abstract Background: Granulomas, hallmark cellular structures of tuberculosis (TB), have similarities to lung tumors, creating a diagnostic and mechanistic interface between infection and malignancy. However, the comparative, genomewide transcriptional landscape between lung cancer and TB granulomas remains poorly characterized. We hypothesize that the lesions in pulmonary TB and cancer may encode convergent molecular programs that regulate the structure and/or function of cells within the lesions. Methods: We performed a large-scale integrative meta-analysis of 450 RNA-seq samples from patients with lung cancer, TB patients, and healthy controls. Raw FASTQ files were processed to generate robust count matrices. Differential gene expression (DGE) was assessed using DESeq2, and co-expression networks were constructed via WGCNA with batch correction. To capture complex, non-linear relationships between genes, pathways, and clinical metadata, we developed a Heterogeneous Graph Transformer (HGT), a graph neural network that models multi-type nodes and edges, enabling the discovery of shared and disease-specific regulatory hubs. Results: Heatmap analysis of the top 100 differentially expressed genes revealed distinct clusters with mixed modules indicative of convergent biology between lung cancer and TB granulomas. KEGG enrichment analysis revealed an overlap between immune and oncogenic signaling pathways, including NF-κB, PI3K-Akt, MAPK, Toll-like receptor, and PD-1/PD-L1 checkpoint pathways. Chromatin remodeling emerged as a common theme, with recurrent histone variants (H2AC, H3C, H4C) suggesting epigenetic plasticity. These shared hub genes implicate chronic inflammation and immune mimicry as mechanistic bridges between granulomas and cancer, making them potential biomarkers to distinguish TB vs lung cancer and/or therapeutic targeting. Conclusions: This systems-level approach reveals a unique TB granuloma-lung cancer interface rooted in common mechanistic pathways, identifying candidate biomarkers and therapeutic targets for two of the world’s most burdensome pulmonary diseases. Citation Format: Andrew Triska, Selvakumar Subbian, Natarajan Ganesan. Heterogeneous graph neural network meta-analysis of lung cancer and tuberculosis transcriptomic datasets reveals convergent host response networks [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 4193.