Multi-omics facilitated variable selection in Cox-regression model for cancer prognosis prediction

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• High-throughput omics technologies enable diverse biomarker measurements.
• Integrating omics data for cancer prognosis prediction remains challenging.
• Existing supervised learning tools for prognosis prediction are limited.

Purpose of the Study:

• To develop novel computational methods for cancer prognosis prediction using integrated omics data.
• To incorporate relationships among different omics datatypes for improved predictive power.

Main Methods:

• Developed SKI-Cox and wLASSO-Cox methods, both fitting the Cox proportional hazards model.
• SKI-Cox uses additional omics data to guide variable selection.
• wLASSO-Cox incorporates omics data relationships as a penalty factor.

Main Results:

• SKI-Cox and wLASSO-Cox models select more true variables than LASSO-Cox in simulations.
• Methods were validated using TCGA glioblastoma and lung adenocarcinoma data.
• Integrated mRNA expression, methylation, and copy number variation data for survival prediction.

Conclusions:

• SKI-Cox and wLASSO-Cox demonstrate superior performance in predicting cancer patient survival.
• These methods offer a more powerful approach to integrating multi-omics data for prognosis.
• The developed tools address the shortage of computational methods for supervised learning in cancer prognosis.