Multiomics-Based Feature Extraction and Selection for the Prediction of Lung Cancer Survival

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Lung cancer poses a significant global health burden, often diagnosed late.
• Accurate prediction of patient survival is crucial for effective treatment strategies.
• Multi-omics data offers a comprehensive view of cancer's molecular complexity.

Purpose of the Study:

• To improve the accuracy of lung cancer survival prediction.
• To develop novel feature extraction and selection methods for multi-omics data.
• To identify robust molecular features associated with patient survival.

Main Methods:

• Utilized gene expression, methylation, and mutation data from TCGA and CPTAC-3 lung adenocarcinoma cohorts.
• Applied gene set aggregation as a feature extraction technique for mutation and copy number variation data.
• Developed and validated machine learning models for 2-year survival prediction using selected molecular features.

Main Results:

• Identified 32 key molecular features from TCGA data, achieving an Area Under the Curve (AUC) of 0.839 for a 2-year survival prediction model.
• Validated the model on the independent CPTAC-3 dataset, yielding an AUC of 0.815 via nested cross-validation.
• Demonstrated the robustness and generalizability of the identified features and prediction model.

Conclusions:

• The integration of multi-omics data with advanced feature engineering significantly enhances lung cancer survival prediction accuracy.
• The identified molecular features provide a strong basis for developing more precise prognostic tools.
• This approach holds promise for improving clinical decision-making and patient management in lung cancer care.