AI and multiomics reveal how gut bacteria trigger colon cancer

Over the past two decades, high‑throughput sequencing has dramatically expanded our understanding of the gut microbiome's composition. However, most studies have simply listed which bacteria are present in colorectal cancer (CRC) patients, without explaining how they actually function. The field now faces major analytical hurdles: microbiome data are compositional (a change in one microbe forces apparent changes in all others), extremely sparse (most bacteria are absent from most samples), and high‑dimensional (thousands of features for a small number of patients). These challenges often lead to spurious correlations and irreproducible results. Because of these problems, a deeper, mechanism‑driven investigation into how microbes truly interact with their human host is urgently needed.

Researchers from the Institute of Digestive Disease at The Chinese University of Hong Kong, led by Professor Jun Yu and Dr. Yinghong Lu, published (DOI: 10.20892/j.issn.2095-3941.2025.0762) their review in Cancer Biology & Medicine. The article systematically dissects how the gut microbiota and host cells talk to each other across four molecular layers - genome, transcriptome, epigenome, and metabolome - and highlights the computational innovations that make such multi‑omics integration possible.

One standout example is the bacterium Escherichia coli carrying the pks island. It produces colibactin, a genotoxin that creates a tell‑tale DNA damage signature found in more than 12% of CRC cases - a direct molecular fingerprint of a microbe causing cancer mutations. Another villain, Fusobacterium nucleatum, uses its virulence factor FadA to latch onto E‑cadherin on host cells, switching on Wnt/β‑catenin signaling and fueling uncontrolled proliferation. The review also spotlights metabolites: secondary bile acids like deoxycholic acid (DCA), generated by certain bacteria, suppress cytotoxic CD8⁺ T cells, helping tumors evade immune attack.

Beyond individual bugs, the authors tackle the computational crisis in microbiome research. They explain how compositional artifacts can fake correlations and how machine learning methods - random forests, neural networks like MetaNN - are now cutting through the noise. Emerging technologies such as long‑read sequencing (PacBio single‑molecule real‑time sequencing and Oxford Nanopore Technologies) and bacterial single‑cell spatial transcriptomics (bacterial multiplexed error‑robust fluorescence in situ hybridization) are also featured. These tools can map exactly which bacterial cells sit next to which host cells inside a tumor, revealing micronicheswhere bacteria drive inflammation or metastasis.

The authors said that the gut microbiome is not a passive passenger in colorectal cancer but an active trigger that rewires host biology from the inside out. "For years we have been looking at lists of bacteria without understanding their real job inside the tumor," they explained. "Now, by combining multi‑omics with AI‑driven models, we can finally see how specific microbes break DNA, silence tumor suppressors, and reprogram the immune system." They added that the biggest challenge is moving from correlations to causation - but with new tools like organ‑on‑chip systems and gnotobiotic mice, that goal is finally within reach.

The insights point directly toward clinical action. Microbiome signatures could one day be used to screen for CRC earlier or to predict who will respond to immunotherapy. Eliminating harmful bacteria - for example, using phages against enterotoxigenic Bacteroides fragilis - may restore chemosensitivity. On the flip side, engineering beneficial bacteria to deliver anticancer payloads or to restore barrier function offers a living therapeutic strategy. The authors envision "digital twins" that integrate a patient's multi‑omics data to forecast how dietary changes, prebiotics, or live biotherapeutics will reshape their individual gut ecosystem. Such precision microbiome medicine could transform CRC prevention, diagnosis, and treatment.

Source:

Chinese Academy of Sciences

Journal reference: