AI based multiomics integration for cancer diagnosis and prognosis

Area of Science:

• Computational biology
• Bioinformatics
• Artificial Intelligence in Oncology

Background:

• Cancer presents a significant global health burden, necessitating advanced diagnostic and therapeutic strategies.
• Integrating diverse biological data (multiomics) holds promise for a more comprehensive understanding of cancer.
• Current approaches may lack the precision required for personalized cancer care.

Purpose of the Study:

• To develop and validate OmicsFusionNet, an AI-powered hybrid model for cancer diagnosis and treatment.
• To assess the model's accuracy across multiple cancer types using integrated multiomics data.
• To explore the utility of OmicsFusionNet in specific applications like ovarian cancer staging.

Main Methods:

• Developed OmicsFusionNet, a hybrid AI model combining machine learning and deep learning techniques.
• Integrated up to six types of multiomics data, including genomics, transcriptomics, and epigenomics.
• Utilized datasets such as CPTAC-OV and TCGA-OV for ovarian cancer analysis and KEGG pathway enrichment for biomarker identification.

Main Results:

• OmicsFusionNet achieved 80.2% accuracy across 23 cancer types.
• Integration of RNAseq and methylation data reached 99.8% accuracy, with XGBoost and deep learning contributing significantly.
• Ovarian cancer staging accuracy ranged from 83% to 91% using combined analytical methods.
• Identified key cancer-related pathways through multiomics biomarker enrichment analysis.

Conclusions:

• OmicsFusionNet demonstrates significant potential for revolutionizing cancer care through AI and multiomics integration.
• The model facilitates precise cancer interventions, biomarker discovery, and personalized treatment strategies.
• This approach advances early cancer detection and uncovers novel mechanisms, ultimately improving patient outcomes.