SFWN: A Novel Semi-Supervised Feature Weighted Neural Network for Gene Data Feature Learning and Mining With Graph...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Genomics

Background:

• Gene expression data analysis is crucial for understanding biological processes.
• Labeling large gene expression datasets is challenging for deep learning models.
• Existing graph neural networks (GNNs) may have limitations in accuracy due to focusing solely on feature and sample spaces.

Purpose of the Study:

• To develop a novel semi-supervised graph neural network model (SFWN) for gene expression data analysis.
• To address the challenge of insufficient labeled data in deep learning for genomics.
• To improve the accuracy and representation ability of gene expression classification.

Main Methods:

• Constructed feature graphs, similarity kernels, and sample graphs using external gene expression knowledge.
• Proposed a novel semi-supervised learning algorithm (SGA) for enhanced data relationship extraction and global sample structure analysis.
• Introduced a graph sparse convolutional network (SGCN) module for sparse representation and classification.
• Implemented a novel feature calculation method to overcome the over-smoothing problem in deep learning models.

Main Results:

• The SFWN model achieved superior performance compared to state-of-the-art approaches.
• Demonstrated high accuracy (0.9993) and F1-Score (0.9899) in gene expression classification.
• Experimental results confirmed the model's strong gene expression feature learning and representation capabilities.

Conclusions:

• The proposed SFWN model effectively handles challenges in gene expression data analysis.
• SFWN offers a powerful new tool for disease diagnosis and clinical practice.
• The model provides new insights into gene expression feature learning and representation.