MARLIN

Methylation- and AI-guided Rapid Leukemia Subtype Inference

MARLIN is a deep neural network model for DNA methylation-based Acute Leukemia classification from sparse DNA methylation profiles.

For more information please refer to our paper.

Installation

Get the repository:

git clone https://github.com/hovestadt/MARLIN

Download MARLIN (trained model) and the human genome assembly hg19

Place them inside the folder files

hg38 and t2t human genome assemblies are now supported, please find probes coordinates in files and change the reference accordingly.

Requirements

R (4.1.3)

R package dependencies: keras (2.13.0) data.table (1.14.2) doParallel (1.0.17) foreach (1.5.2)

Conda environment setup:

conda create --name marlin -c conda-forge r-base=4.1.3
conda activate marlin
conda install -c conda-forge r-keras=2.13 r-tensorflow=2.13 tensorflow-gpu=2.13 python=3.10

The official Oxford Nanopore Technologies tool to extract DNA modifications modkit

Real-time classification

MARLIN can be used to generate methylation class predictions in real-time during live basecalling. Real-time script waits for bam files and it processes them as they are produced. The files are expected to be from the same sample and they are processed cumulatively.

For details: go to the real-time folder

shinyMARLIN (currently in beta version!)

The webapp shinyMARLIN allows users to upload genome-wide methylation calls to generate Acute Leukemia methylation class predictions.

Learn more about shinyMARLIN

Training

Usage (specific CUDA device 1): CUDA_VISIBLE_DEVICES=1 Rscript MARLIN_training.R

Prediction

Input bed file format: chromosome, start, end, methylation call (0 to 1 or NA for not covered), probe name (e.g. cg21870274)

Reference CpGs

Usage (specific CUDA device 1): CUDA_VISIBLE_DEVICES=1 Rscript MARLIN_prediction.R