# MPCI: A novel metric for quantifying DNA methylation patterns in NGS data

## TL;DR

MPCI is a new method to better detect and analyze DNA methylation patterns in sequencing data, improving disease detection and liquid biopsy accuracy.

## Contribution

MPCI introduces a novel metric for capturing consistent methylation patterns across sequencing reads, outperforming existing methods in distinguishing methylation profiles.

## Key findings

- MPCI outperforms existing metrics in distinguishing closely related cell types and multi-tissue classification.
- MPCI detects in-silico cfDNA spike-ins at as low as 1% abundance and performs better in clinical liquid-biopsy data.
- MPCI achieves higher classification accuracy in pre- vs. post-transplant cfDNA profiles compared to dMHL.

## Abstract

Epigenetic processes, particularly disruptions in DNA methylation profiles, are associated with many disease states. Traditional approaches for DNA methylation biomarker discovery focusing on individual CpG sites do not account for fragment-level methylation states. Methylation haplotype analysis offers a more comprehensive approach leading to increased distinction capability between reads originating from tissues with diverse methylation profiles. This can be particularly valuable in liquid biopsy where detecting small amounts of disease-specific cell-free DNA (cfDNA) amidst a bulk of healthy cfDNA is challenging. To address limitations of existing metrics for quantifying methylation patterns in a region from sequencing data, we propose the Methylation Pattern Consistency Index (MPCI), a novel metric that captures consistent methylation patterns across sequencing reads, accounting for both methylated and unmethylated blocks of CpGs. Using whole-genome bisulfite sequencing data, we demonstrate that MPCI outperforms MHL and its symmetric counterpart, dMHL (MHL – uMHL), across several benchmarks: distinguishing closely related cell types (CD4 vs. CD8; AUC 0.915), multi-tissue classification (0.92 accuracy), and detection of in-silico cfDNA spike-ins at abundances as low as 1%. Notably, in a clinical liquid-biopsy cohort of liver transplant patients, MPCI achieved significantly higher classification performance than dMHL (Accuracy: MPCI: 0.868 ± 0.023 vs. dMHL: 0.768 ± 0.027, p = 0.014) in discriminating pre- from post-transplant cfDNA profiles. These findings position MPCI as a reliable quantification approach for biomarker selection or diagnostic testing in epigenetic studies. We have made MPCI available as an R function for usage convenience.

DNA methylation is a chemical tag added to DNA that helps control how genes are turned on and off. Disruptions to these patterns are common in diseases such as cancer, and traces of these changes can be detected in the small fragments of DNA circulating in blood—a technique known as liquid biopsy. A major challenge, however, is that disease-derived signals are often overwhelmed by abundant DNA from healthy cells.

We developed a new computational approach, the Methylation Pattern Consistency Index (MPCI), which identifies consistent methylation patterns—both present and absent—across neighboring DNA sites in sequencing data. We show that MPCI outperforms existing methods by more effectively distinguishing closely related cell types, detecting tissue-derived DNA at levels as low as 1%, and robustly tracking changes in real patient liquid biopsy samples.

MPCI provides researchers and clinicians with a more sensitive and balanced way to quantify epigenetic signals. This approach has the potential to improve early disease detection, enhance treatment monitoring, and increase the accuracy of liquid biopsy–based diagnostics across a wide range of conditions.