MONTE

MONTE: Methylation-based Observation Normalization and Tumor purity Estimation

Note

Slow git clone speed: If you experience slow download speeds, this is because we've included all pretrained models in this repository. In a future official release, we plan to move the pretrained models to cloud storage to improve download performance. Thank you for your patience.

---

Quick naviagation

• MONTE
  • Quick naviagation
  • Installation
  • Instruction and usages
    • Input data
    • Training the model
    • Using pretrained models
    • Purity esitmation
    • Methylation purification
    • Bayesian transfer learning
    • Save and load
  • Analysis
  • Citation

Installation

# download the repo
git clone https://github.com/ylaboratory/MONTE.git
cd MONTE

# you can install it in your python environment via your favorite package manager
conda activate your_env_name
# mamba activate your_env_name

# in the MONTE folder
pip install .

Instruction and usages

Input data

There are two types of input data required for model training: beta values and purity scores.

• Beta Values:
  The beta values should be stored in a pandas DataFrame with the shape (samples, probes). It is essential that the column names (i.e., probe names) are provided, as they will be used to align probe features when applying the model to a new dataset. If your new dataset contains a different set of probes, the model will automatically compute the intersection of the probe sets.

• Purity Scores: The purity scores should be provided as a pandas Series.

Note

Using a Pretrained Model: If you are using a pretrained model, only the beta values are required.

Training the model

To train your own model, the Monte provides a similar interface like the scikit-learn, which creating a model object and fit / fit_transform with you data. The following scripts shows how to train a model, except the input data beta values, and purity

import torch
from monte import Monte

# Create the model
# user can use default values directly
model = Monte()

# Assume your beta value table is X, and the y is your purity scores
model.fit(X, y)

Using pretrained models

Another way to use this package is through the pretrained models, which allow users to skip the training process. This package provides 22 pretrained models which trained on the 450K methylation microarray from the TCGA.

To view the available pretrained models, you can use available_pretrained_models(). And you can further access the model by from_pretrained(model_name).

from monte import available_pretrained_models, from_pretrained

# print all pretrained models
print(available_pretrained_models())

# access to the cancer-specific model
model = from_pretrained("COAD")

# access to the pan-cancer model
pan_cancer_model = from_pretrained("PAN-CANCER")

The model works identically to the manually trained models, all the functions are avaiable.

Purity esitmation

To estimate tumor purity scores, either train your own model or use a pretrained model. Then, pass your input beta values (or M-values) matrix to the predict_purity function.

estimated_purity = model.predict_purity(your_data)

Methylation purification

To purifiy the methlyation values, the purify_values() function will internally esitmate the purity first and then adjust the values by given target purity you want. In default the target purity is 1.

purified_beta = model.purify_values(your_data, target_purity=1)

Bayesian transfer learning

The MONTE package also provides functionality to leverage existing or pretrained models on new datasets with different purity metrics. Use the fine_tune function to adapt the model:

model.fine_tune(your_data, new_purity_scores)

This updates the model coefficients and internal statistics to smoothly adapt to your new dataset.

Save and load

To reuse the model, the package provides functionality to easily save and load MONTE models.

# save model
model.save(your_model_saved_path)

# load model
# since loading a model requires a Monte instance, you can load it using the class method
model = Monte.load(your_model_saved_path)

Analysis

---

Citation

The preprint is available on bioRxiv.

If you use MONTE in your research, please cite our preprint:

@article {Kim2026.01.22.701164,
author = {Kim, Mirae and Lee, Wei-Hao and Yao, Vicky},
title = {MONTE: Methylation-based Observation Normalization and Tumor purity Estimation},
elocation-id = {2026.01.22.701164},
year = {2026},
doi = {10.64898/2026.01.22.701164},
publisher = {Cold Spring Harbor Laboratory},
URL = {https://www.biorxiv.org/content/early/2026/01/23/2026.01.22.701164},
eprint = {https://www.biorxiv.org/content/early/2026/01/23/2026.01.22.701164.full.pdf},
journal = {bioRxiv}
}