Protocol for non-invasive tumor monitoring and diagnosis based on interpretable deep learning

TL;DR

This paper introduces a non-invasive method using deep learning to track tumor DNA methylation in blood samples for cancer monitoring and diagnosis.

Contribution

A new protocol using the interpretable deep-learning framework Oncoder to monitor tumor treatment response via cfDNA methylation.

Findings

Oncoder tracks dynamic changes in tumor-specific DNA methylation signals in plasma cfDNA.

The protocol includes steps for differential methylation analysis and model training with tumor-specific markers.

The method allows statistical comparison of tumor fraction dynamics after drug therapy.

Abstract

Tumor-specific DNA methylation profiling in plasma cell-free DNA (cfDNA) offers a promising approach for non-invasive tumor detection. Here, we present a protocol that uses Oncoder, an interpretable deep-learning-based framework, to monitor treatment response by tracking dynamic changes in tumor-specific DNA methylation signals in patient plasma cfDNA. We describe steps for data preparation, performing differential methylation analysis, training Oncoder, and interpreting the model’s outputs. This protocol is versatile and adaptable to various data types and application scenarios.

For complete details on the use and execution of this protocol, please refer to Yang et al.1

•A protocol for using Oncoder to predict tumor fractions from cfDNA methylation data•Instructions for differential DNA methylation analysis•Steps for training the Oncoder model using tumor-specific DNA methylation markers•Statistical comparison of dynamics in tumor fractions following drug therapy

A protocol for using Oncoder to predict tumor fractions from cfDNA methylation data

Instructions for differential DNA methylation analysis

Steps for training the Oncoder model using tumor-specific DNA methylation markers

Statistical comparison of dynamics in tumor fractions following drug therapy

Publisher’s note: Undertaking any experimental protocol requires adherence to local institutional guidelines for laboratory safety and ethics.

Tumor-specific DNA methylation profiling in plasma cell-free DNA (cfDNA) offers a promising approach for non-invasive tumor detection. Here, we present a protocol that uses Oncoder, an interpretable deep-learning-based framework, to monitor treatment response by tracking dynamic changes in tumor-specific DNA methylation signals in patient plasma cfDNA. We describe steps for data preparation, performing differential methylation analysis, training Oncoder, and interpreting the model’s outputs. This protocol is versatile and adaptable to various data types and application scenarios.

Figures