AI-driven discovery of cell-state correctors for age-associated diseases

Correcting the cell states that drive aging and the diseases it causes

Dark Matter Therapeutics has built the Programmable Human Pharmacology Engine (PHPE), an AI platform that models how a molecule changes the state of diseased cells and predicts how that response plays out across the human body.

Explore the Platform

The Challenge

Drug discovery keeps asking the wrong question.

The old question

Does this molecule bind the target?

Target-based drug discovery

The new question

Does this molecule correct diseased cell biology in the human body?

In vivo cell-state-based drug discovery

Where discovery breaks

One target, not the disease.

Drug discovery addresses a single protein rather than the diseased biology itself.

PHPE models the whole diseased cell state, not a single target.

Models run out of data.

Predictions lose reliability once biology or chemistry moves past what has been measured.

PHPE is built to predict beyond its training data.

Cells in a dish are not patients.

What works in vitro rarely holds up in people.

PHPE is built to translate predictions from cells to patients.

From cell state to human response

Aging is not wear-and-tear. It is a coordinated shift in how cells function. PHPE reads the whole diseased cell state, designs molecules to move it back toward health, and predicts how those molecules behave in the human body. It is target-agnostic and designed for multi-target biology.

PHPE models the chain from molecule to protein network to cell state to clinical pharmacology. It builds on methods that generate novel single compounds or combinations to revert a diseased cell state to a healthy one with optimized PK profiles and deconvoluted molecular targets.

Schematic of cell-state correction Three rows of gene-level changes. Healthy cells sit near the baseline, diseased cells shift many genes up or down, and with a cell-state corrector most shifts shrink back toward the baseline. Healthy cells Diseased cells Diseased cells + cell-state corrector
Schematic, not data. Each bar is a gene shifted up or down from its healthy level. Disease shifts many at once; a cell-state corrector moves them back toward the healthy pattern.
How PHPE works

Inputs

  • Human omics data
  • Chemistry data
  • Real-world patient data
  • New chemotypes, as structures only

PHPE

  • Machine learning
  • Physics-based modeling
  • Systems biology
  • Knowledge graphs

Predicted human response

  • Cell-state correction
  • Tissue exposure (PK)
  • Target engagement
  • Biomarkers
  • Whole-body safety and toxicity

Lab in a loop. Only the top-ranked molecules are made and tested in human cells, for target engagement and in vivo, and the results retrain PHPE.

Predict Response

From a molecule’s structure alone, predicts how it shifts cell state, along with its tissue exposure (PK), target engagement and biomarkers, before it is made.

Identify Toxicity

Flags whole-body safety and toxicity risks early, so weak candidates are removed before costly studies.

Optimize Molecules

Generates new chemotypes and optimizes them for efficacy and safety together. Only the top-ranked candidates are synthesized for testing.

Bridge the Gap

Closes the loop between patients, in silico models, cells and in vivo studies, so each experiment improves the next prediction.

Beyond the data

Existing biological and chemical data cover only a small part of what matters. Informed by first principles of physics and systems biology, PHPE is built to generalize beyond what has been measured, to compounds, cell types and exposures outside its training data: the “dark” biology that current datasets do not cover.

Matching drugs to patients

PHPE is designed to find drugs for biomarker-defined patient groups that are large and underserved. In published work, the founders’ MultiDCP model predicted drug responses for individual Alzheimer’s patients from their own gene-expression profiles, ranking repurposing candidates for each patient on efficacy and toxicity.

Evidence

Early proof of concept in Alzheimer’s disease

Transcriptomics, human iPSC-derived neurons

In human cells

A PHPE-predicted compound (DMTx0005) partially reversed the disease-associated gene biomarkers, including SCGN and IFITM2, in human iPSC-derived cortical neurons carrying familial Alzheimer’s mutations (PSEN2 N141I or APP V717I).

Behavior and histology, TgF344-AD rats

In animals

In female TgF344-AD rats, a model of Alzheimer’s disease, DMTx0005 reduced spatial-learning deficits and microglial activation. Males showed no cognitive benefit.

Active-comparator claims cohort

In patients’ records

In a US insurance-claims cohort of 364,733 people, exposure to a drug flagged by the platform was associated with 57–62% lower incidence of Alzheimer’s than an active antihistamine comparator (hazard ratios 0.38–0.43).

Our Thesis

The mechanisms of aging underlie many common diseases

Shared molecular drivers of aging converge on distinct clinical phenotypes. We correct the cell states they create, aiming to prevent or modify disease rather than treat symptoms.

Mechanistic drivers of aging

  • Mitochondrial dysfunction
  • Cellular senescence
  • Metabolic dysfunction
  • Inflammation
  • Epigenetic alteration
  • Genomic instability
  • Telomere shortening
  • Hormonal regulation

Our programs

  • Alzheimer’s disease Globally $8B to $13B in 2034
  • Kidney fibrosis Globally $12B to $25B in 2034

Pipeline

Program Signature ID Lead discovery Lead optimization Development candidate IND-enabling
Alzheimer’s disease Early efficacy biomarkers: blood Aβ42/Aβ40, p-tau, NfL; PET imaging
Now: lead discovery
Kidney fibrosis Early efficacy biomarkers: blood PIIINP; urine MCP-1, KIM-1, DKK-3
Now: signature identification

Each program has blood, urine or imaging biomarkers that could give an early read on efficacy in clinical trials.

Founding team and advisors

Founders

Steven Zimmer

Steven Zimmer

Co-Founder & CEO

Scientist and serial entrepreneur. Co-founded Protein Sciences Corp., acquired by Sanofi in 2017. Founder and CEO of EpiCombi.AI.

Prof. Lei Xie

Prof. Lei Xie, Ph.D.

Scientific Founder & CSO

Professor of Pharmaceutical Sciences and Computer Science at Northeastern University. Creator of the core AI algorithms behind the platform. Former researcher at Roche.

You Wu

You Wu, Ph.D.

Co-Founder

Ph.D. in Computer Science, City University of New York. Developed the core machine-learning algorithms and leads the development of PHPE.

Advisors

Prof. Philip E. Bourne

Prof. Philip E. Bourne, Ph.D.(In Memoriam)

Advisor

Founding Dean, School of Data Science, University of Virginia. Former NIH Associate Director, President of ISCB and founding Editor-in-Chief of PLOS Computational Biology.

Dr. Stephen Harrison

Dr. Stephen Harrison

Strategic Advisor

Former CSO and CEO of multiple biotechnology companies in the US and UK. Developed multiple drugs into clinical development and onto the market. Drives wet-lab biology at DMTx.

Thomas Hartung

Thomas Hartung, M.D., Ph.D.

Advisor

Doerenkamp-Zbinden Chair, Johns Hopkins Bloomberg School of Public Health. Pioneering AI and organoid cultures to replace animal testing. Advises on validation strategy and implementation.

Prof. Paul Brennan

Prof. Paul Brennan, Ph.D.

Advisor

Professor, Centre for Medicines Discovery, University of Oxford. Medicinal chemistry expertise; advises on candidate selection.

Help us correct the cell states behind aging

We are looking for investors and pharma partners to take our first programs from lead discovery to a development candidate. Reach the right person directly.

Investment and partnerships

Steven Zimmer

Co-Founder & CEO

Science and research collaborations

Prof. Lei Xie

Scientific Founder & CSO

Email
lxie@iscb.org