DOMSCNet: a deep learning model for the classification of stomach cancer using multi-layer omics data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Next-generation sequencing (NGS) has accelerated biomedical research, particularly in understanding cancer's molecular basis.
• Analyzing complex, high-dimensional multi-layer omics data for cancer is crucial for diagnosis and therapy.
• Existing computational methods struggle with diverse omics data and effective feature extraction for integrated analysis.

Purpose of the Study:

• To develop a robust hybrid feature selection (HFS) technique for optimal feature detection from multi-layer omics datasets.
• To propose a novel hybrid deep recurrent neural network-based model, DOMSCNet, for stomach cancer classification.
• To ensure the model's generalizability across different omics datasets and validate its performance on external datasets.

Main Methods:

• A hybrid feature selection (HFS) technique, including SelectKBest-maximum relevancy minimum redundancy-Boruta (SMB), was developed.
• A novel hybrid deep recurrent neural network model, DOMSCNet, was designed for stomach cancer classification.
• The DOMSCNet model was trained and validated on four multi-layer omics datasets and eight external datasets.

Main Results:

• The proposed SMB HFS technique demonstrated superior performance compared to other HFS methods.
• The DOMSCNet model achieved higher accuracy in classifying stomach cancer across multiple omics datasets.
• DOMSCNet outperformed existing and other proposed classifiers in validation tests.

Conclusions:

• The developed HFS technique effectively extracts optimal features from complex omics data.
• DOMSCNet offers a powerful and generalizable deep learning approach for stomach cancer classification.
• This study advances computational methods for multi-layer omics data analysis in cancer research.