Mapping MASH: Understanding Liver Disease from Blood Samples

As new therapies emerge for metabolic dysfunction-associated steatohepatitis (MASH), the need for scalable and precise diagnostics has never been greater. To address this demand, a partnership between medical centers and biotechnology company Hepta has created a comprehensive, multiomic atlas of MASH. This collaboration connects liver tissue biology with circulating cfDNA methylation signatures, enabling the measurement of fibrosis, inflammation, and metabolic dysregulation directly from blood.

In this Innovation Spotlight, Hamed Amini, the cofounder and chief executive officer of Hepta, explains the need for improved liver disease diagnostics and describes how the company’s AI-based liquid biopsy analysis platform can provide accurate diagnostics from noninvasive patient samples.

What is MASH, and why is now the time for better diagnostics?

MASH, formerly known as non-alcoholic steatohepatitis, is the progressive, inflammatory form of metabolic dysfunction–associated steatotic liver disease (previously referred to as non-alcoholic fatty liver disease). It affects more than 20 million Americans, yet fewer than 1 in 10 are diagnosed today.^1,2 For two decades, clinicians have relied on tools that cannot scale: biopsies that are invasive and impractical, imaging devices such as FibroScan that are unavailable in most primary care settings, and blood scores such as Fibrosis-4 (FIB-4) that produce false positives.

Meanwhile, the MASH treatment landscape has changed. With Rezdiffra approved in 2024, Wegovy approved in 2025, a half dozen therapies in Phase 3 clinical trials, and more than a dozen in Phase 2, physicians finally have more options.³ What they lack is a way to identify the right patients early enough to benefit. The bottleneck is no longer therapy, but rather detection. We now need diagnostics that are biologically rigorous, scalable, and compatible with everyday clinical workflows. That is the gap Hepta is addressing.

How did the collaboration for the MASH dataset come about, and what makes this atlas unique?

Our scientific strategy has always centered on anchoring liquid biopsy signals to real tissue biology. To do that, we formed collaborations with leading hepatology centers at Duke University, Mainz University, University of California, San Diego, as well as pharmaceutical industry partners including Akero Therapeutics’ Phase 3 program. Together, we built an unparalleled paired cfDNA–liver-tissue multiomic dataset for MASH. The atlas integrates single-cell and bulk gene expression, chromatin accessibility, and DNA methylation across dozens of human liver samples, all paired with matched plasma cfDNA methylation from blood, alongside hundreds of additional biopsy-confirmed plasma cfDNA samples. This dataset uniquely maps how methylation orchestrates pathway activity across hepatocytes, cholangiocytes, and stromal compartments, and shows that those same pathway signatures are detectable in plasma. It creates a molecular bridge between tissue-level biology and plasma cfDNA—a bridge between organ solid biopsy and blood liquid biopsy.

What is the AI technology that made this project possible?

Our platform uses the first liquid-biopsy-native AI platform, which gives rise to the unique transformer model optimized specifically for cfDNA, called LiquidTransformer. Rather than interrogating small genomic regions or relying on hand-designed features, the model processes every observed cfDNA fragment in context. A single blood sample can contain up to a billion unique DNA fragments. Our patented architecture uses the same attention mechanism that underpins large language models, but it is devised for and scaled to molecular data.

This allows the model to evaluate how every fragment relates to every other fragment, revealing subtle, distributed, and nuanced methylation patterns that are completely missed by fragment-based or region-based approaches developed for oncology. Chronic diseases such as MASH do not produce loud, focal genomic signals; they reshape the epigenome in millions of small ways. Detecting those shifts requires a model with full contextual awareness.

This transformer architecture is what makes it possible to decode biology from a blood draw at a depth that historically required a biopsy.

What enables cfDNA methylation to capture pathway-level biology occurring in organs?

When organs undergo metabolic stress, inflammation, or fibrogenesis, those changes are encoded in the epigenetic programs that regulate transcription. Methylation determines which genes are turned up or down, and it changes in a coordinated way as the disease progresses.

As hepatocytes and other liver cell types turn over, fragments of their DNA enter the bloodstream. Those fragments carry the methylation marks that reflect pathway activity inside the tissue. What our atlas shows is that key biological signatures, metabolic dysfunction, inflammatory activation, extracellular matrix remodeling, and bile-acid pathway dysregulation are tightly correlated between liver tissue and circulating cfDNA.

The consistency is mechanistic. cfDNA carries the organ’s regulatory fingerprint. When you pair that with a transformer trained to interpret billions of fragments in context, you can read those pathway-level programs directly from blood.

What insights from the atlas have stood out the most so far?

First, the atlas maps coordinated methylation and transcriptional programs that tightly track fibrosis severity. These include metabolic reprogramming, inflammatory cascades, fibrogenic pathways, and mechanisms targeted by emerging drug classes, such as glucagon-like peptide-1 (GLP-1), fibroblast growth factor 21 (FGF-21), and thyroid hormone receptor beta (THR-β) agonists. The ability to observe these therapeutic mechanism-aligned signatures in cfDNA opens the door to more targeted therapy selection.

Second, across independent cohorts from Duke University, Mainz University, and Akero Therapeutics, we see reproducible detection of the same cfDNA methylation signatures. That robustness across populations is critical for clinical translation.

Third, the cfDNA signal is not just a yes-or-no diagnostic. It reflects dynamic biological programs, including the molecular fingerprint of failed repair. This provides a potential window into prognosis, molecular staging, and treatment response.

The atlas shows that cfDNA methylation functions as a faithful mirror of liver biology, with far richer information than traditional fibrosis biomarkers.

How will scientists apply the knowledge gained from the MASH atlas?

The atlas is designed as a foundational resource for the field, enabling advances across diagnostics, therapy selection, drug development, prognostic modeling, and mechanistic research. It anchors liquid-biopsy biomarkers to tissue-level mechanisms, helps identify molecular subtypes of MASH that align with specific drug mechanisms, and supports the discovery of new therapeutic targets by tracing dysregulated pathways and cell-state transitions. It also links epigenetic signatures to disease trajectory and outcomes, providing a basis for more accurate prognostic models, while revealing how hepatocyte, cholangiocyte, immune, and stromal programs interact to drive fibrosis. Because it connects methylation, gene expression, and cfDNA for each sample, the atlas creates an integrated map of MASH biology that allows researchers to link molecular mechanisms directly to clinically measurable readouts.

What is next for Hepta’s AI platform?

The platform is moving from proof to scale. Clinically, we are expanding validation cohorts, preparing for regulatory pathways, and working with hepatologists to bring this test into specialist and primary-care settings. Scientifically, we are developing models for therapy guidance and longitudinal monitoring, analogous to minimal residual disease monitoring in oncology but tuned for chronic disease. The same cfDNA framework can track the fractional presence of disease-specific signals over time, offering a new way to quantify disease burden and treatment response. Because the technology reads the full epigenetic landscape, not just liver-specific features, we see a path toward detecting other chronic diseases using the same assay and architecture. Our long-term vision is a platform that provides a multi-organ view of health from a single blood draw.