Home

Welcome to the MethylNet wiki!

Levy, J. J., Titus, A. J., Petersen, C. L., Chen, Y., Salas, L. A., & Christensen, B. C. (2019). MethylNet: A Modular Deep Learning Approach to Methylation Prediction. BioRxiv, 692665. https://doi.org/10.1101/692665

This convenient command line tool was designed for deep learning biological discovery in the methylation space, a tool that could be scaled to high-throughput workflows in the future.

The goal of this wiki is to build upon concepts introduced in https://github.com/Christensen-Lab-Dartmouth/PyMethylProcess/wiki , and show how you can conduct the following deep learning tasks on methylation data. The package comes with many options for each, so we'll try to explain what is happening:

1. Embeddings
2. Predictions (regression, classification)
3. Hyperparameter Optimization
4. Model Interpretation (Clustering of Embeddings, SHAP feature attributions)

The MethylNet API documentation and usage of all of its classes and functions can be found here: https://christensen-lab-dartmouth.github.io/MethylNet/

Please take a look at the example scripts for usage of how to run the complete pipeline.

We're going to continue the analysis of our sample dataset (GSE87571) demonstrated in https://github.com/Christensen-Lab-Dartmouth/PyMethylProcess/wiki , so as a prerequisite please read through that wiki before continuing.

Note: Alternatively, a quick way to generate a MethylationArray object is to make 2 Pandas DataFrames, one for the beta values (rows are sample names, columns cpgs), and one for the pheno data (rows sample names), create a python dictionary with keys "beta" and "pheno", store the dataframes with those keys, and then serialize the dictionary using pickle.