The GEM package provides tools for simulating data and fitting a GEM model using Torch in R. It includes functions to generate synthetic datasets and fit a model based on those data.

Install with:

renv::install("weinstockj/GEM")

Use simulate_data to generate a synthetic dataset:

This function returns a list containing:

• x: a numeric matrix of predictors with one row for each candidate somatic mutation and one column for each feature
• y: a numeric response vector containing zcore'd age at blood draw
• z: an indicator for whether a candidate mutation is a "real"
• sample_map: a list mapping samples to indices
• true_count and naive_count for model assessment

Fit the GEM model on your data using the fit_model function:

model <- fit_model(x, y, sample_map, lr = 0.1, iters = 180, num_threads = 4L)

The returned list includes:

• Model coefficients (e.g. beta),
• Fitted mutation burden estimates (fitted),
• Loss history, and other parameters.

Please see the "articles" portion of the package website for a brief vignette.

To run the tests for GEM, execute:

devtools::test()

Please cite our manuscript if you use this package in your research.