← Research
Research project 2026 Preprint

Recovering tumor-intrinsic DNA methylation signals from bulk cancer epigenomes

MONTE estimates tumor purity and corrects DNA methylation at single-CpG resolution, recovering tumor-intrinsic regulatory signals from bulk cancer samples.

Labels
Cancer epigenomics DNA methylation Tumor purity
MONTE workflow showing pan-cancer model training, Bayesian transfer learning, tumor-purity estimation, methylation correction, and downstream applications
MONTE learns pan-cancer relationships between DNA methylation and tumor purity, adapts them to new datasets through Bayesian transfer learning, and produces purity-adjusted methylation profiles for downstream analysis.

Research question: How can we separate tumor-intrinsic methylation patterns from the non-tumor signal mixed into bulk cancer samples?

The biological problem

DNA methylation profiling is a cornerstone of cancer epigenomics, but a bulk tumor measurement combines signals from tumor and non-tumor cells. That mixture can obscure the regulatory patterns that belong to the tumor itself and can confound downstream biological or clinical analyses.

The MONTE framework

MONTE is a cancer-agnostic framework that estimates tumor purity and corrects DNA methylation signals directly from bulk data. It works at single-CpG resolution by learning probe-wise relationships between observed methylation and tumor purity.

The method does not require matched normal samples, predefined probe sets, or cancer-specific reference models. This makes the same framework applicable across cancer types while keeping the correction interpretable at the level of individual methylation probes.

Why it matters

By recovering tumor-intrinsic epigenomic profiles, MONTE provides cleaner inputs for studying cancer regulation and for downstream analyses that would otherwise inherit tumor-purity bias.

The preprint describes the model, validation, and cross-cancer evaluation. The implementation is also listed on the research software page.

ARUNA, developed in collaboration with Janmajay and the Yao laboratory, addresses a complementary methylation problem. It imputes missing or undetected methylation profiles from a subset of probes and can upscale lower-resolution RRBS measurements toward higher-resolution WGBS profiles.