Estimating gene expression from DNA methylation and copy number variation: A deep learning regression model for...

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Gene expression analysis is crucial for understanding cancer biology.
• Multi-omics data offers a comprehensive view of cancer development but integration is challenging.
• A robust model to predict gene expression from multi-omics data is currently lacking.

Purpose of the Study:

• To develop a deep learning-based predictive model for estimating gene expression from genetic and epigenetic data.
• To quantitatively capture the correlation between genetic/epigenetic alterations and gene expression directionality in liver hepatocellular carcinoma (LIHC).

Main Methods:

• Utilized a deep learning approach combining Deep Denoising Auto-encoder (DDAE) and Multi-layer Perceptron (MLP).
• DDAE was employed to extract significant features from multi-omics data.
• MLP was used for regression and classification tasks, learning from the extracted features.

Main Results:

• The developed model quantitatively correlates genetic and epigenetic alterations with gene expression.
• Benchmarked against state-of-the-art regression models.
• Achieved 95.1% accuracy in disease classification for LIHC.

Conclusions:

• The proposed deep learning model effectively integrates multi-omics data for gene expression prediction.
• This approach provides a robust method for understanding molecular insights in cancer.
• The model demonstrates high accuracy in classifying liver hepatocellular carcinoma.