CancerNet

CancerNet classifies cancer tissue of origin from epigenomic profiles using a Variational Autoencoder (VAE)-based deep learning model trained on The Cancer Genome Atlas (TCGA) encompassing 33 cancer types.

Key Features:

• VAE-based deep learning architecture: Uses a Variational Autoencoder (VAE)-based deep learning model to represent and model complex epigenomic patterns.
• Unified cancer model: Integrates data from 33 distinct cancer types as represented in TCGA to enable pan-cancer analysis.
• Epigenomic feature exploitation: Leverages dysregulated epigenomes and hypomethylation patterns as early molecular indicators of carcinogenesis.
• Distinguishes cancer stages and types: Differentiates pre-cancerous lesions, primary cancers, metastatic tumors, second primary cancers, and cancers of unknown primary.
• High classification accuracy: Achieves an overall F-measure greater than 99% for tissue-of-origin classification across diverse cancers.
• Early detection capabilities: Characterizes pre-cancer samples to support early detection of cancerous changes.

Scientific Applications:

• Cancer diagnostics: Assigns tissue of origin to support diagnostic classification of cancer samples.
• Research on cancer origins: Analyzes epigenomic changes to study early carcinogenesis and molecular distinctions among cancer types.
• Metastatic and secondary cancer detection: Identifies tissues of origin for metastatic tumors, second primary cancers, and cancers of unknown primary.

Methodology:

CancerNet applies a VAE-based deep learning approach trained on TCGA epigenomic data to learn patterns in dysregulated epigenomes that are indicative of cancer types and stages.