Decoding tumour bacterial ecosystems: Topological data analysis of bacterial association with immunogenicity.

Abstract

10563 Background: The tumour microbiome is increasingly recognised as a key contributor to cancer progression and clinical outcomes, as highlighted in recent iterations of the Hallmarks of Cancer. Translating microbial signatures into actionable insights is hampered by a reliance on dimensionality reduction and a limited capacity to dissect non-linear relationships, leading to incomplete and inconsistent findings. Topological Data Analysis (TDA) is an unsupervised machine learning method that overcomes these limitations by preserving the complexity of high-dimensional data. Here, we applied TDA to characterise tumoural bacterial ecosystems across cancer types. Methods: Analysis was conducted using BioCorteX’s knowledge graph and proprietary engines v20250124_015926. The dataset constituted of 551 sequenced primary tumour microbiome samples from six cancer types, as well as associated host age and gender. TDA Mapper was used to generate network graphs of the bacterial data, combined with an enrichment algorithm to analyse host metadata. Clusters in the TDA network represent samples with overlapping microbiome features. Results: The resultant network graph shows clear clustering of tumour bacteria along cancer types, suggesting distinct signatures of tumoural bacteria for each cancer. Notably, TNCB and melanoma cluster in the same area, while ovarian, colorectal, and glioblastoma cluster together in a different area of the graph. NSCLC clusters separately to both of the other two major clusters. Host age and gender do not significantly interact with the bacterial signatures or the cancer types. Conclusions: This strengthens the notion of highly cancer-specific tumoural bacteria, while also highlighting the similarities observed across cancer types. Notably, the 2 main clusters could suggest a bacterial association with immunogenicity: TNBC and melanoma tumours are both immune-responsive, while ovarian, colorectal, and glioblastoma tumours are generally not. Previous studies have suggested a major role of tumoural bacteria in immune responses, and this study suggests that tumoural bacteria can distinguish cancer immunogenicity. Further, it demonstrates the potential of TDA to extract novel insights from large, high-dimensional datasets, surpassing traditional approaches. Capturing non-linear associations is crucial for understanding complex host-bacteria interactions, paving the way for more precise characterisation of cancer-specific microbial signatures, with implications for oncology and therapeutic development.