methCancer-gen

methCancer-gen generates cancer-type-specific DNA methylome datasets using deep learning and a conditional variational autoencoder (CVAE) to enable study of cancer-associated DNA methylation patterns.

Key Features:

• Conditional Variational Autoencoder (CVAE): Employs a CVAE neural network-based generative model to estimate the conditional distribution of DNA methylation data with latent variables.
• Training on Existing Methylome Data: Learns to capture the complex distribution of methylation patterns by training on existing DNA methylome datasets.
• User-Specified Cancer-Type Generation: Generates methylome samples conditioned on user-defined cancer types to produce cancer-specific datasets.
• Benchmark Evaluation and High Accuracy: Has been evaluated against benchmark methods and shown the capability to reproduce cancer type-specific methylome datasets with high accuracy.

Scientific Applications:

• Explore Epigenetic Mechanisms: Enables analysis of generated methylation signatures to investigate how DNA methylation influences gene activity in different cancers.
• Facilitate Biomarker Discovery: Provides simulated methylome datasets to aid identification of methylation-based biomarkers for early detection and prognosis.
• Support Drug Development: Supplies cancer-specific methylation data for investigating epigenetic targets relevant to therapeutic intervention.

Methodology:

Trains a conditional variational autoencoder (CVAE) using existing DNA methylome data within a deep learning framework and evaluates generated datasets against benchmark methods to assess accuracy and preservation of statistical properties.