SALMON: Survival Analysis Learning With Multi-Omics Neural Networks on Breast Cancer

Area of Science:

• Computational Biology
• Genomics
• Precision Medicine

Background:

• Accurate cancer prognosis is crucial for precision medicine.
• Existing models struggle to aggregate complex multi-omics data for survival prediction.
• Sophisticated algorithms are needed to identify key predictors and data types for improved prognosis.

Purpose of the Study:

• To implement Deep Learning networks for predicting breast cancer survival using gene expression data.
• To develop an algorithm (SALMON) that aggregates and simplifies multi-omics data for prognosis prediction.
• To assess the impact of different omics data types on predictive performance.

Main Methods:

• Utilized Deep Learning-based neural networks for survival analysis.
• Developed the Survival Analysis Learning with Multi-Omics Neural Networks (SALMON) algorithm.
• Employed eigengene modules from gene co-expression network analysis as model inputs, alongside feature selection and enrichment analysis.

Main Results:

• Demonstrated improved prediction performance with the inclusion of more omics data.
• Successfully used eigengene modules, rather than raw gene expression, for enhanced model interpretation.
• Identified biologically relevant co-expression modules associated with breast cancer prognosis.

Conclusions:

• Deep Learning models, particularly SALMON, show feasibility for breast cancer survival analysis.
• Integrating multi-omics data and gene co-expression modules enhances prognostic prediction.
• This approach offers a blueprint for future Deep Learning-based survival analyses in oncology.