Colorectal Cancer Prediction Based on Weighted Gene Co-Expression Network Analysis and Variational Auto-Encoder

Area of Science:

• Genomics
• Bioinformatics
• Cancer Research

Background:

• Accurate prediction models are crucial for effective cancer diagnosis.
• Colorectal cancer (CRC) diagnosis relies on analyzing complex genomic data.
• Gene expression data offers potential biomarkers for cancer prediction.

Purpose of the Study:

• To develop and evaluate an effective two-step feature extraction method for colorectal cancer (CRC) prediction.
• To identify key genes and reduced dimensional representations for improved CRC classification.
• To assess the performance of a support vector machines (SVM) classifier using these novel features.

Main Methods:

• Microarray gene expression data for 238 colorectal cancer (CRC) and normal samples were analyzed.
• Weighted Gene Co-expression Network Analysis (WGCNA) identified 12 gene modules, with a key module highly correlated to CRC.
• Hub genes were screened from the key module, and variational autoencoder (VAE) was used for dimensionality reduction of differentially expressed genes.

Main Results:

• Ten hub genes were selected as one type of feature.
• A 10-dimensional representation was generated using VAE for another set of features.
• The support vector machines (SVM) classifier achieved high accuracy (0.9692) and Area Under the Curve (AUC) (0.9981) using the combined features.

Conclusions:

• The proposed two-step feature extraction method, integrating WGCNA-derived hub genes and VAE-based dimensionality reduction, demonstrates high efficacy for CRC prediction.
• This approach offers a promising strategy for enhancing the accuracy of cancer prediction models.
• The findings highlight the potential of combining network-based gene analysis with deep learning for biomarker discovery in oncology.