Hybrid deep learning framework for accurate classification of high dimensional genomic data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-dimensional genomic data presents challenges for deep learning (DL) due to noise and sparsity.
• Existing DL models struggle with interpretability and stability on heterogeneous genomic structures.

Purpose of the Study:

• To develop a novel hybrid deep learning framework for improved classification of high-dimensional genomic data.
• To enhance feature selection and adaptive refinement for better accuracy and interpretability.

Main Methods:

• A hybrid TabNet-CNN framework integrating attention-driven feature selection and convolutional refinement was proposed.
• The TabNet component utilizes attention mechanisms to identify salient genomic features.
• Convolutional Neural Network (CNN) layers refine localized feature interactions for precise classification boundaries.

Main Results:

• The proposed TabNet-CNN framework demonstrated superior performance over existing state-of-the-art models.
• Significant improvements were observed in classification accuracy, Area Under the Curve (AUC), and model interpretability.
• The framework effectively handles noisy, redundant, and sparse features in genomic datasets.

Conclusions:

• The hybrid TabNet-CNN framework offers a robust solution for analyzing complex genomic data.
• The model's enhanced interpretability and performance are promising for precision medicine.
• Potential applications include biomarker discovery, disease subtyping, and clinical decision support.