DeepMetabio-mCRC Screener: A Multi-Omics Deep Learning Framework for Early Risk Prediction and Biomarker Discovery in Colorectal Liver Metastasis

  • Comput Struct Biotechnol J. 2026 May 25;35(1):0074. doi: 10.34133/csbj.0074.
Hongyu Zhang  1 ,  Ke Wang  2 ,  Runqiu Guo  1 ,  Xiaochuan Wu  3 ,  Qingquan Chen  2 ,  Qiaojun He  1 ,  Bo Yang  1  4 ,  Yanyan Zhuang  2 ,  Wanling Yang  5 ,  Hong Zhu  1
Affiliations
  • 1. Zhejiang Province Key Laboratory of Anti-Cancer Drug Research, College of Pharmaceutical Science, Innovation Institute for Artificial Intelligence in Medicine of Zhejiang University, Zhejiang University, Hangzhou 310058, China.
  • 2. Guangdong Provincial Key Laboratory of Malignant Tumor Epigenetics and Gene Regulation, Sun Yat-sen Memorial Hospital, Sun Yat-sen University, Guangzhou 510120, China.
  • 3. State Key Laboratory of Advanced Drug Delivery and Release Systems, Institute of Pharmaceutics, College of Pharmaceutical Sciences, Zhejiang University, Hangzhou 310058, China.
  • 4. School of Medicine, Hangzhou City University, Hangzhou, 310015, China.
  • 5. Department of Paediatrics and Adolescent Medicine, The University of Hong Kong, Hong Kong, China.
PMID: 42199376 DOI: 10.34133/csbj.0074
Abstract

Colorectal liver metastasis (CRLM) remains the primary cause of mortality in patients with colorectal Cancer (CRC), yet effective predictive tools and reliable biomarkers are still lacking. DeepMetabio-mCRC Screener, an integrated multi-omics framework combining large-scale transcriptomic profiles with serum metabolomics, was developed to address this gap. In a cohort of 1,077 CRC samples, 620 metabolism-related genes were used to train a convolutional neural network, yielding an area under the receiver operating characteristic curve of 0.92 in the validation cohort and 0.97 in the independent testing cohort, outperforming the performance of the 10 established machine learning models. Model-derived transcriptomic risk scores revealed 22 core metabolic features associated with metastatic progression and CRLM occurrence, particularly retinol and tryptophan metabolism. Cross-omics integration revealed aminocarboxymuconate-semialdehyde decarboxylase (ACMSD) as a promising biomarker associated with impaired nicotinamide adenine dinucleotide biosynthesis. Clinical validation in 100 CRC patients confirmed elevated ACMSD levels in patients with CRLM, which correlated with advanced stage, recurrence risk, an immune-inflamed tumor microenvironment, and heightened sensitivity to epidermal growth factor receptor/vascular endothelial growth factor receptor-targeted therapies. In vitro, ACMSD knockdown was associated not only with suppressed CRC cell migration caused by inhibition of the transforming growth factor-β/epithelial-to-mesenchymal transition pathway but also with decreased proinflammatory and immune-responsive pathways and reduced immune cell infiltration. These findings collectively validate the DeepMetabio-mCRC Screener as a substantial early risk prediction tool and underscore ACMSD, identified through this framework, as a multifunctional biomarker for diagnosis, prognosis, molecular characterization, and therapeutic decision-making in patients with CRLM.

Products