MethylNet

MethylNet applies deep learning to DNA methylation (DNAm) data to produce embeddings, predictive models, and synthetic methylation profiles for studying cellular heterogeneity, cancer subtypes, aging, and environmental exposures such as smoking.

Key Features:

• Automated deep learning framework: Constructs embeddings, trains predictive models, and generates new DNAm data for downstream analysis.
• High-dimensional DNAm handling: Models continuous, interacting, and nonlinear methylation data to mitigate issues such as multiple hypothesis testing and multicollinearity.
• Integration with Pythonic data types: Operates on Pythonic MethylationArray data structures to manage methylation datasets.
• PyTorch-based implementation: Leverages the PyTorch framework for model building and training.
• Disease heterogeneity capture: Learns complex nonlinear interactions relevant to cellular differences and cancer subtype characterization.
• Epigenetic analysis capabilities: Enables biological age estimation and detection of environmental exposure–associated methylation signals such as smoking.

Scientific Applications:

• Cellular differences: Studies variation between cell types by capturing cell-type–specific methylation patterns.
• Cancer sub-type analysis: Extracts higher-order methylation information to support characterization of cancer subtypes.
• Aging processes: Estimates biological age from DNAm data.
• Environmental exposure detection: Identifies methylation patterns associated with exposures such as smoking.

Methodology:

Implements deep learning models using PyTorch to construct embeddings, train predictive models, and generate methylation data while operating on Pythonic MethylationArray data structures.