Hepta, a liquid biopsy company focused on liver disease, today announced research demonstrating how its LiquidTransformer technology reads liver biology from DNA circulating in blood. A paper accepted to the 2026 Conference on Neural Information Processing Systems (NeurIPS) establishes why chronic disease requires a different AI architecture from approaches developed for cancer, and how Hepta’s LiquidTransformer solves that problem. Applying it to liver disease, Hepta unveiled six studies at The Liver Meeting demonstrating the technology’s capabilities across diagnosis, drug research and disease characterization. This includes a study with Madrigal focused on identifying patients eligible for approved MASH treatments, in which Hepta’s test performance matched magnetic resonance elastography (MRE), the most accurate non-invasive method and one largely confined to specialist centers and clinical trials, from a routine blood draw.

MASH, or metabolic dysfunction-associated steatohepatitis, is a progressive liver disease involving inflammation and scarring that can advance to cirrhosis and liver failure. Approved therapies target patients with moderate-to-advanced fibrosis, stages F2 to F3, so the extent of a patient’s scarring determines who qualifies for the approved drugs. Establishing that stage today means an invasive liver biopsy, or an imaging that most patients cannot easily reach. Hepta’s first commercial product in development is a blood test designed to identify those patients.

A scarring liver can release far more DNA into blood than an early-stage tumor, yet tests reading this circulating cell-free DNA (cfDNA) have not entered clinical practice for diagnosing liver fibrosis or MASH. Tumors can leave distinctive abnormalities concentrated in particular genomic regions. Chronic liver disease, however, produces much subtler changes spread across the genome as liver cells lose function, inflammation develops and scar tissue forms. LiquidTransformer was built to read those changes together.

“Think of cancer as a pinch of black pepper in a bucket of salt. There’s very little of it, but it stands out, so the job is to spot the specks. Chronic liver disease is a spoonful of sugar in that bucket. There’s far more of it, but it looks just like the salt, so the tools that spot pepper can’t see it. You need a different way of looking, and that’s what we built LiquidTransformer to do. We are making the liver as readable from blood as a tumor is today, with our first commercial test identifying patients who qualify for MASH treatment. Since the technology is organ- and disease-agnostic, our ambition is to expand this capability to other organs and indications beyond the liver,” said Hamed Amini, Ph.D., co-founder and CEO of Hepta.

Three of the six studies presented at The Liver Meeting 2026, hosted by the American Association for the Study of Liver Diseases (AASLD) in Denver, were conducted with leading MASH drug developers. Madrigal Pharmaceuticals, maker of Rezdiffra, the first approved MASH therapy, collaborated on diagnostic validation of LiquidTransformer on an independent cohort. Inventiva, whose pan-PPAR candidate Lanifibranor is in Phase 3, collaborated on the biology underlying its therapeutic approach. Akero Therapeutics, a Novo Nordisk company, collaborated on research supporting Hepta’s targeted diagnostic assay.

A new AI architecture for liquid biopsy in chronic disease recovers liver biology from blood

Hepta’s paper, accepted to the NeurIPS 2026 main track, “Why Cancer cfDNA Models Fail on Chronic Disease: A Geometric Information-Theoretic Bound and Its Architectural Implications,” establishes a mathematical framework explaining why AI architecture determines whether a model can recover the faint, widely distributed signals of chronic disease. The analysis, spanning 9,432 tissue samples across 14 cancer types and 26 chronic diseases, shows that simply adding more data or building bigger models on the old architectures cannot close the gap.

The same architecture, LiquidTransformer, captures the faint methylation patterns across hundreds of millions of cfDNA fragments at once. These chemical markings carry information about the identity and biological state of the cells that released them. Paired with Hepta’s curated liver disease atlas, the architecture connects what it reads in blood to what is happening inside the liver.

A study selected for oral presentation at AASLD’s The Liver Meeting by Anna Mae Diehl, M.D., of Duke University tested that connection directly. Researchers measured biological changes in liver tissue, then examined whether LiquidTransformer could recover those changes from blood. LiquidTransformer identified 17 of the top 20 disease-related pathway changes which conventional blood cell-free DNA analysis largely missed. As fibrosis advanced, the blood readout also detected liver cells losing their characteristic identity and bile-duct cell populations expanding (Abstract 0205).

In an independent cohort of 229 biopsy-confirmed participants from Madrigal’s MASH program, LiquidTransformer matched MRE for identifying treatment-eligible fibrosis and outperformed the other evaluated tests for that task. MRE is the most accurate non-invasive method for staging liver fibrosis, but it requires an MRI scanner equipped with elastography hardware and software, trained technologists and expert interpretation, which concentrates its availability in specialist centers and clinical trials. Hepta matched its performance for the treatment-eligibility decision from an ordinary blood draw. Additionally, LiquidTransformer outperformed every other test evaluated, including FibroScan, Pro-C3, FAST and MAST in head-to-head comparison. Its advantage is held at the boundary between mild and significant fibrosis, where the treatment decision falls and conventional tests lose accuracy (Abstract 2303: Poster of Distinction).

“For years, MRE imaging has been the non-invasive benchmark we rely on in our research. It’s the most accurate way we have to stage liver fibrosis without a biopsy. Seeing a blood test approach the diagnostic accuracy of MRE and beat VCTE and ELF at determining treatment eligibility is a real step forward, as that is an important unmet need given the wider availability and access of a blood test. Further studies are underway to validate these findings,” said Rohit Loomba, M.D., of UC San Diego, Hepta’s Chief Medical and Scientific Advisor.

A separate study advanced the targeted assay intended for Hepta’s first commercial test. Across three cohorts, including Akero’s SYNCHRONY program, a panel covering less than 1% of the genome retained the diagnostic performance of whole-methylome analysis, supporting the cost-effective deep sequencing a practical test requires (Abstract 2322).

Early studies reveal drug action and differences across liver diseases beyond MASLD/MASH

In collaboration with Inventiva, researchers analyzed Hepta’s MASH atlas and identified coordinated methylation changes across the PPAR family, which regulates liver metabolism, inflammation and scarring. Different family members were associated with different aspects of disease, supporting the biological rationale for therapies that engage multiple PPAR isoforms (Abstract 1052).

In the 52-week SAMARA trial of 35 participants receiving semaglutide or placebo, LiquidTransformer detected molecular changes consistent with drug activity even though an MRI-based response measure did not distinguish the treatment groups. Semaglutide is the molecule behind the medicines sold as Wegovy and Ozempic. Liver-cell programs became more active, while bile-duct programs associated with scarring were suppressed. Responders also began with molecular patterns that treatment appeared to reverse. (Abstract 3485).

In a study of 60 participants, the same technology read all three forms of steatotic liver disease and told them apart: MASLD, driven by metabolic dysfunction; ALD, driven by alcohol; and MetALD, where the two coexist. MetALD, the newest category, carried its own signature of immune activation and scarring rather than sitting midway between the other two. The blood readout also captured the liver’s response to alcohol-related injury. Clinicians today ask patients how much they drink, and PEth blood tests measure recent intake, but neither shows what alcohol has done to the liver. The findings point toward one blood test across the full spectrum of steatotic liver disease (Abstract 3004: Poster of Distinction and ALD Special Interest Group highlights presentation).

Presentations at The Liver Meeting 2026

The meeting takes place November 5 to 9 in Denver.

  • Abstract 0205. Plasma cfDNA Methylation Enables Non-Invasive Snapshot of Liver Pathway Activity, Cell Composition, and Fibrosis Stage. Presenter: Anna Mae Diehl, M.D., Duke University. Oral presentation, November 7.

  • Abstract 1052. Coordinated methylation across PPAR isoforms in plasma cfDNA reflects pathway-wide epigenetic dysregulation in MASLD and supports a pan-PPAR therapeutic rationale. Presenter: Manal Abdelmalek, M.D., Mayo Clinic. Poster, November 5.

  • Abstract 2303. External Validation of LiquidTransformer in Madrigal MASH Program: Hepta’s AI-powered Cell-Free DNA Methylation Biomarker Outperforms All Evaluated Blood-Based and Elastography Markers for Detecting Treatment-Eligible MASH. Presenter: Rohit Loomba, M.D., UC San Diego. Poster of Distinction, November 6.

  • Abstract 2322. Enrichment Panel Designed for the Study of Chronic Liver Disease: Targeted Methylation cfDNA Assay Enables Cost-effective Ultra-deep Sequencing for Identification of Treatment-Eligible MASH Patients. Presenter: Jörn Schattenberg, M.D., Saarland University. Poster, November 6.

  • Abstract 3004. Plasma cell-free DNA methylation resolves three molecular subtypes of steatotic liver disease (SLD) and reveals MetALD as biologically distinct from a MASLD-to-ALD continuum. Presenter: Rohit Loomba, M.D., UC San Diego. Highlights talk, November 5; Poster of Distinction, November 7.

  • Abstract 3485. Plasma cell-free DNA methylation recovers semaglutide pharmacology in MASH trials despite an uninformative MRI-PDFF endpoint. Presenter: Rohit Loomba, M.D., UC San Diego. Poster, November 7.

All four presenters at The Liver Meeting serve as advisors to Hepta. Additionally, the paper titled “Why Cancer cfDNA Models Fail on Chronic Disease: A Geometric Information-Theoretic Bound and Its Architectural Implications” is authored by Fereshteh Abedini and Soheil Damangir of Hepta and will be presented at the 40th edition of NeurIPS, which takes place December 6-13 across Sydney, Atlanta, and Paris.

About Hepta

Hepta reads liver biology from a blood draw. Its LiquidTransformer AI platform combines analysis of methylation patterns across circulating DNA fragments with a curated liver disease atlas to recover information about disease processes and treatment-related changes. Its first diagnostic in development is designed to identify patients eligible for MASH treatment, with additional research applications spanning drug action and the spectrum of steatotic liver disease. The architecture also supports future investigation in other organs and chronic diseases. Based in Foster City, California, Hepta’s founders bring experience from Illumina, GRAIL and Google. Investors include Felicis Ventures, Illumina Ventures, SeaX Ventures, Alumni Ventures and AME Cloud Ventures. Visit hepta.bio.

Media gallery

About The Author
Business Wire