Deep neural network-based biostatistical analysis for disease marker screening.

Area of Science:

• Computational biology
• Biostatistics
• Machine learning in healthcare

Background:

• Traditional statistical methods struggle with high-dimensional, small-sample biological data.
• Deep learning methods, particularly deep neural networks (DNNs), show promise for complex data analysis.
• Biomarker screening is crucial for disease diagnosis and treatment.

Purpose of the Study:

• To propose and evaluate a novel deep neural network (DNN) framework for biomarker screening.
• To compare the DNN framework's performance against traditional methods like LASSO and random forests.
• To enhance the interpretability of DNN models for clinical application.

Main Methods:

• Development of a deep neural network (DNN) framework for biomarker screening.
• Integration of an attention mechanism and SHapley Additive exPlanations (SHAP) for model interpretability.
• Comparative experiments using a breast cancer dataset and validation on single-cell sequencing data.

Main Results:

• DNN models significantly outperformed traditional methods in sensitivity, accuracy, and Area Under the Curve (AUC) for biomarker screening.
• The combined attention mechanism and SHAP analysis provided guided biological interpretations.
• The framework demonstrated scalability for multi-omics data integration and enhanced explanatory power.

Conclusions:

• The proposed DNN framework offers a powerful and interpretable approach to biomarker screening, outperforming traditional statistical methods.
• The model's interpretability features facilitate clinical understanding and application.
• This framework shows potential for multi-omics data integration, cross-disease applications, and advancing precision medicine.