EMT network-based feature selection improves prognosis prediction in lung adenocarcinoma

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Reproducibility of cancer prognostic biomarker discovery is a challenge.
• High dimensionality and dataset variability impact feature selection performance.
• Integrating biological networks and multi-omics data are key for robust predictive models.

Purpose of the Study:

• To develop and validate a phenotype-relevant network-based feature selection (PRNFS) framework.
• To improve lung cancer prognosis prediction using epithelial mesenchymal transition (EMT) relevant networks and multi-omics data.
• To assess the performance of single- and multi-omics EMT prognostic signatures.

Main Methods:

• Constructed cancer prognosis-relevant networks based on epithelial mesenchymal transition (EMT).
• Integrated biological networks with multi-omics data for feature selection using the PRNFS framework.
• Evaluated prediction performance using Area Under the Curve (AUC) and sample stratification.

Main Results:

• Achieved remarkable prediction performance (average AUC >0.8) with less than 2.5% of total dimensionality.
• Identified EMT prognostic signatures with significant sample stratifications and biological interpretability.
• Multi-omics signatures significantly improved sample stratification compared to single-omics signatures.
• Validated findings on independent multi-omics lung cancer datasets.

Conclusions:

• The PRNFS framework enhances the discovery of robust cancer prognostic biomarkers.
• EMT signatures derived from integrated omics data offer significant improvements in lung cancer prognosis prediction and patient stratification.
• Network-based integration of multi-omics data is a promising strategy for biomarker discovery.