Enhancing cancer subtype classification through convolutional neural networks: a deepinsight analysis of TCGA gene...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional gene expression data presents challenges for accurate cancer subtype classification.
• Existing computational models may not fully capture complex patterns within genomic data.

Purpose of the Study:

• To adapt and evaluate DeepInsight for multi-class cancer subtype classification using gene expression data.
• To compare DeepInsight's performance against established machine learning models.
• To identify key genes associated with different cancer subtypes.

Main Methods:

• DeepInsight was adapted to process gene expression data by converting it into image representations.
• Performance was evaluated against support vector machines, LightGBM, neural networks, and decision trees.
• A novel multi-class feature selection method using aggregated class activation maps was developed.
• Gene Ontology analysis was performed on identified critical genes.

Main Results:

• DeepInsight consistently achieved higher F1 scores than traditional models across breast, lung, and colon cancer datasets.
• The method successfully addressed multi-class classification challenges.
• Several significant genes were identified across multiple classification methods.
• Gene Ontology analysis revealed the biological roles of critical genes.

Conclusions:

• Adapted DeepInsight offers a novel approach for cancer subtype classification by leveraging image-based representations of gene expression data.
• Aggregated class activation maps effectively pinpoint critical features for gene discovery.
• DeepInsight shows promise as a valuable tool for both cancer subtype classification and the identification of biologically relevant genes.