Integrating Clinical and Multiple Omics Data for Prognostic Assessment across Human Cancers

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Limited comprehensive assessment of prognostic values of multi-omic profiles across diverse cancer types.
• Need for systematic evaluation of genomic, epigenomic, and transcriptomic data in cancer prognosis.

Purpose of the Study:

• To conduct a pan-cancer prognostic assessment using a multi-omic kernel machine learning approach.
• To systematically quantify the prognostic values of different omic profiles individually and integratively.
• To evaluate the combined prognostic value of omics and clinical factors.

Main Methods:

• Utilized a multi-omic kernel machine learning method for prognostic assessment.
• Analyzed 3,382 samples across 14 cancer types, integrating genomic, epigenomic, and transcriptomic data.
• Compared prognostic performance of individual omic profiles, integrated omics, and omics combined with clinical factors.

Main Results:

• Prognostic performance varied significantly across cancer types, with mRNA and miRNA expression profiles showing the best performance.
• DNA methylation profiles also demonstrated strong prognostic value.
• Germline susceptibility variants consistently showed low prognostic performance.
• Integration of omic profiles with clinical variables substantially improved prognostic performance in half of the cancers studied.
• The kernel machine learning method outperformed existing prognostic signatures.

Conclusions:

• Omic data, particularly mRNA, miRNA, and DNA methylation, holds significant prognostic value in cancer.
• Integrating omics with clinical data offers a powerful approach for refined cancer prognosis.
• Genome-wide omic biomarker aggregation via kernel machine learning enhances prognostic assessment for precision oncology.