TabDEG: Classifying differentially expressed genes from RNA-seq data based on feature extraction and deep learning...

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Traditional methods for identifying differentially expressed genes (DEGs) struggle with small sample sizes due to distribution assumptions, leading to high error rates.
• Deep learning (DL) offers a promising alternative for analyzing gene expression data, but challenges remain in labeling and sample size for RNA-Seq data.
• Data augmentation (DA) can generate valuable pseudo-values from limited data, enhancing feature extraction without substantial cost.

Purpose of the Study:

• To develop a robust model, TabDEG, integrating Data Augmentation (DA) with a Deep Learning (DL) framework for improved DEG identification.
• To accurately predict DEGs and their regulatory directions (up-regulation/down-regulation) from gene expression data.
• To address the limitations of traditional models in high-dimensional, small sample size datasets, particularly in cancer genomics.

Main Methods:

• Proposed TabDEG model combining DA and DL-based tabular data modeling.
• Utilized gene expression data from The Cancer Genome Atlas (TCGA) database.
• Compared TabDEG performance against five existing DEG identification methods.

Main Results:

• TabDEG demonstrated high sensitivity and low misclassification rates compared to counterpart methods.
• The model effectively enhances data features for classifying high-dimensional, small sample size datasets.
• Predicted DEGs from TabDEG significantly mapped to important gene ontology terms and cancer-associated pathways.

Conclusions:

• TabDEG is a robust and effective method for identifying DEGs in challenging small sample size datasets.
• The integration of DA and DL provides a powerful approach for analyzing RNA-Seq data in cancer research.
• TabDEG facilitates the discovery of biologically relevant genes and pathways implicated in cancer development.