P2409 - Meta-Analysis of Multi-Omics Data Combined With AI for Personalized Therapeutics in Colorectal Cancer

Mohamed Farghaly, MBBCh¹, Yara Mohamed, MBBCh², Alaa Abdelsalhin, MBBCh³, Moatazbellah Mustafa, MBBCh⁴, Omar Sabry, MBBCh⁵, Mariam Mohamed, MBBCh², Sara Elsenary, MBBCh⁵, Ahmed Elsayed, MBBCh⁵, Mahmoud M. Elsayed, MD⁶
1New Giza University, 6th of October, Al Jizah, Egypt; 2New Giza University, Rehab City, Al Qahirah, Egypt; 3Misr University for Science and Technology, Rehab City, Al Qahirah, Egypt; 4Modern University for Technology & Information, Rehab City, Al Qahirah, Egypt; 5New Giza University, New Cairo, Al Qahirah, Egypt; 6MME Foundation, Monsoura, Ad Daqahliyah, Egypt

Introduction:

Colorectal cancer (CRC) remains a major cause of cancer mortality globally. While therapeutic advancements exist, a "one-size-fits-all" approach limits optimal outcomes. Multi-omics (genomics, transcriptomics, proteomics) integration offers a holistic view of tumor biology. When coupled with artificial intelligence (AI), it enables stratification of patients and identification of novel therapeutic targets—ushering in an era of precision oncology.

To conduct a meta-analysis of studies utilizing AI models on multi-omics data to identify clinically actionable biomarkers and predictive signatures for personalized CRC therapeutics.

 

Methods:

We conducted a systematic search (2010–2024) across PubMed, Scopus, and Embase in adherence with PRISMA guidelines. Eligible studies applied machine learning (ML) or deep learning (DL) techniques to multi-omics datasets for CRC prognostication or therapeutic response prediction. Using R (meta, metafor) and Python (scikit-learn, lifelines), we extracted and pooled AUROC, accuracy, and hazard ratios (HR) of models predicting therapy response. Quality assessment was performed using PROBAST and QUADAS-2.

Results:

A total of 26 studies encompassing 5,473 CRC patients were included. Integrated AI models trained on multi-omics data achieved a pooled AUROC of 0.89 (95% CI: 0.85–0.93) for predicting therapeutic response. DL approaches (especially convolutional neural networks) outperformed traditional ML (p < 0.01). Pathway enrichment identified consistent involvement of Wnt/β-catenin, TGF-β, and PI3K/AKT signaling. AI-derived biomarkers (e.g., MSI status, CMS subtype, CD8+ T-cell infiltration) aligned strongly with survival benefit (HR = 0.42; p = 0.003) under immunotherapy. Few models were externally validated, underscoring the need for real-world clinical trials.

Discussion:

AI applied to multi-omics data significantly enhances prediction of CRC treatment response and survival, revealing opportunities for individualized therapy selection. Our meta-analysis supports the clinical potential of AI-omics integration and highlights the need for validation pipelines. This approach may redefine CRC management by tailoring therapies to each tumor’s molecular blueprint.

Disclosures:


Mohamed Farghaly, MBBCh¹, Yara Mohamed, MBBCh², Alaa Abdelsalhin, MBBCh³, Moatazbellah Mustafa, MBBCh⁴, Omar Sabry, MBBCh⁵, Mariam Mohamed, MBBCh², Sara Elsenary, MBBCh⁵, Ahmed Elsayed, MBBCh⁵, Mahmoud M. Elsayed, MD⁶. P2409 - Meta-Analysis of Multi-Omics Data Combined With AI for Personalized Therapeutics in Colorectal Cancer, ACG 2025 Annual Scientific Meeting Abstracts. Phoenix, AZ: American College of Gastroenterology.