DEGnext: classification of differentially expressed genes from RNA-seq data using a convolutional neural network with...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Traditional differential expression analysis struggles with small sample sizes, leading to potential inaccuracies.
• Deep learning (DL) offers promise for RNA-seq data analysis but faces challenges like limited labeled data and small sample sizes.
• Transfer learning enhances DL model performance on new datasets by leveraging patterns from related data.

Purpose of the Study:

• To develop a novel Convolutional Neural Network (CNN) model for predicting upregulating (UR) and downregulating (DR) genes from RNA-seq data.
• To apply transfer learning to a CNN model for improved biomarker prediction in both trained and untrained cancer datasets.
• To address limitations in current differential gene expression analysis for small datasets.

Main Methods:

• Implemented DEGnext, a CNN model utilizing biologically validated data and logarithmic fold change values.
• Applied transfer learning to DEGnext, enabling the transfer of learned feature maps to new, unseen cancer datasets.
• Compared DEGnext's performance against traditional machine learning methods using Receiver Operating Characteristic (ROC) scores.

Main Results:

• DEGnext achieved competitive performance (ROC scores 88-99%) compared to Decision Tree, K-Nearest Neighbors, Random Forest, Support Vector Machine, and XGBoost.
• The model demonstrated robustness and effectiveness in classifying unseen datasets through transfer learning.
• Predicted differentially expressed genes (DEGs) from DEGnext mapped to significant cancer-related Gene Ontology terms and pathways.

Conclusions:

• DEGnext accurately classifies UR and DR genes from RNA-seq cancer data with high performance.
• The model's use of biologically relevant fine-tuning data aids in biomarker exploration.
• DEGnext's approach can be adapted for biomarker discovery in other disease datasets.