American Chemical Society CRT | Model Medicines Publishes Class-Leading, Regulatory Grade Toxicology Model AmesNet™, Validation of GALILEO™ Platform & ChemPrint™ Model Underpinning Drug Pipeline

SAN DIEGO — Model Medicines, an AI-first biotechnology company developing first-in-class therapeutics against multi-indication biological choke points, today announced that the company’s AmesNet™ agent, a deep learning model for predicting Ames test mutagenicity, has been peer-reviewed and published in the American Chemical Society (ACS) journal, Chemical Research in Toxicology (CRT).

AmesNet™ demonstrated best-in-class performance across all benchmarked approaches in predicting chemical safety on out-of-domain data. The agent outperformed models from the FDA (DeepAmes/NCTR/CDER), MIT (ChemProp), Tencent (GROVER), and the University of Sydney^[1]. The model uses a novel AI architecture called Task-Conditioned Learning (TCL) and is built on GALILEO™, the same AI platform Model Medicines uses to discover and de-risk its drug pipeline.

Key Highlights:

Peer-Reviewed Validation: American Chemical Society (ACS) publication, Chemical Research in Toxicology (CRT), a 1st Quartile CiteScore Journal.

Best-in-Class Performance: Late stage Ames test failures jeopardize an estimated $10 million in capital and multiple years of developmental progress per candidate. AmesNet™ detects compounds that models from the FDA, MIT, Tencent, and the University of Sydney miss. AmesNet recovered 106 more true positives than the FDA’s DeepAmes agent across 13 mutagenic substructure classes. The publicly available benchmark set contained 560 Ames true positives.

New Modeling Paradigm: Task-Conditioned Learning (TCL) architecture captures assay-context-dependent mutagenicity that all prior “unconditioned” models either don’t account for or average away. 

TCL Encoder-Swap: Integrating competitor agents into Model Medicines’ TCL framework improved their performance.

Early-Stage Screening Enabled: AmesNet’s performance enables high-confidence Ames screening at scale during early discovery, pulling preclinical safety assessment forward from a late-stage bottleneck to a continuous input in compound prioritization.

Platform Extension: AmesNet™ is the first public release from Model Medicines' multi-modal suite of ADMET GALILEO™ agents.

Validated Engine Behind the Pipeline: AmesNet™ is part of the end-to-end AI drug discovery and development platform, GALILEO™, which Model Medicines used to discover its two lead assets, MDL-001 and MDL-4102.

Open Science: All data and code (STL, uMTL, gMTL, TCL encoder swaps, and bootstrapping) are available in the following GitHub repository: github.com/Model-Medicines/TCL-Ames.

Publication: The peer-reviewed article is available here: doi.org/10.1021/acs.chemrestox.6c00082.

“Peer-review and publication of the AmesNet™ toxicology agent in the prestigious toxicology journal American Chemical Society CRT underscores the importance and success of our foundational GALILEO™ platform," said Daniel Haders, PhD, CEO of Model Medicines. "AmesNet demonstrates that Model Medicines can predict toxicity with world-class, regulatory grade accuracy in entirely novel chemical spaces - outcompeting the best models in the world on a public dataset. AmesNet was built on our GALILEO platform and utilizes our ChemPrint agent. The same platform and agent discovered our lead, first-in-class assets, MDL-001 and MDL-4102, and the novel broad-spectrum Thumb-1 drug target. The explicit implication is that the GALILEO platform and the ChemPrint agent are one of, if not the most productive, accurate and creative platform and agent in AI Drug Discovery.”

Background: The Ames Testing Bottleneck

The Ames test is a decades-old laboratory assay that uses bacteria to check whether a chemical can cause mutations in DNA. These “mutagens” carry an early warning that they may cause cancer. For that reason, the Ames test is a required genotoxicity assay for all novel small-molecule therapeutics prior to human clinical trials. GLP-compliant testing often exceeds $10,000 per compound and requires approximately 2 grams of material, making routine screening impractical during early discovery. Developers typically defer testing until regulatory submission, at which point tens of millions of dollars have been sunk into programs. Regulatory agencies have moved to enable AI alternatives: the FDA Modernization Act provides a legal framework for computational models to reduce wet-lab testing, the FDA’s AI4TOX/SafetAI initiatives target AI for toxicological endpoints, and ICH M7 endorses computational Ames prediction. No model has met this mandate.

Task-Conditioned Learning (TCL)

The Ames test is a battery of experiments across multiple bacterial strains. Each strain is sensitive to a different type of DNA mutation, conducted with and without a liver enzyme fraction (S9) that simulates human drug metabolism. A compound may be mutagenic in one strain but not another, or only when metabolized. All prior AI models are “Unconditioned” and produce a single prediction without conditioning on strain or metabolic activation. AmesNet’s TCL architecture resolves this issue with a dual-branch design: one branch encodes molecular structure; the second encodes assay conditions (strain identity and ±S9 status). The model learns separate decision boundaries for each context rather than averaging across all conditions.

Correctly identifying mutagenic compounds, known as sensitivity, is the most critical metric in AI-driven Ames testing. False negatives allow dangerous compounds to advance undetected. Existing models fail on sensitivity because compound classes, such as planar aromatic intercalators and aromatic amines, produce context-dependent signals that Unconditioned Models dilute. Structural enrichment analysis confirms AmesNet™ recovers these classes.

AmesNet™ Benchmark Results

AmesNet™ was evaluated on a withheld out-of-domain test set of 4,208 data points comprising compounds chemically dissimilar from the training data:

“Current AI models fail because they try to force complex biological problems into a single prompt," notes Tyler Umansky, Head of Platform and Machine Learning and the paper's lead author. "AmesNet proves the power of doing the opposite: separating every variable, discovering winners and losers for each one, and finding the chemistry that wins for all variables. AmesNet is a microcosm of how we approach all of drug discovery, doing the tedious but necessary hard work. It’s how we built our two lead programs and established a pipeline capable of tackling more than 30 indications.”

A Paradigm Shift for Ames Screening and Preclinical Development

A model with AmesNet’s sensitivity and balanced accuracy in novel chemical space enables high-confidence mutagenicity screening at scale during early drug discovery. This shifts Ames testing from a wet-lab, late-stage gate on a single candidate to a computational filter applied across entire compound libraries at program concept. This repositioning advances preclinical safety assessment by years in the development timeline, and transforms Ames screening from a late-stage binary pass/fail checkpoint into a continuous, scalable input for compound prioritization. AmesNet™ enables developers to eliminate mutagenic liabilities from their pipelines before they become costly.

GALILEO™: One Platform, Discovery Through Development

AmesNet™ is not a standalone agent. It is the latest capability within GALILEO™, the AI platform Model Medicines uses to discover, optimize, and de-risk its own wholly owned drug pipeline. AmesNet™ and the Company’s drug programs share the same molecular foundation: the encoder that powers AmesNet is the same agent used to discover the company’s two lead assets, MDL-001 and MDL-4102.

MDL-001 — a direct-acting, non-nucleoside, broad-spectrum antiviral targeting the conserved viral polymerase mechanism (RdRp Thumb-1^[2]), with demonstrated preclinical activity across respiratory and hepatic viruses and high-risk co-infections. MDL-001 is being developed across major respiratory infections, including influenza, COVID-19, and RSV, as well as chronic hepatitis infections, including HCV, HBV, and HDV^[3], representing an estimated combined global antiviral market exceeding $30 billion annually. Oral MDL-001 demonstrates preclinical equivalency or superiority against established standards of care, specifically oseltamivir (Tamiflu) for influenza, nirmatrelvir (Paxlovid), molnupiravir (Lagevrio), and remdesivir (Veklury) for SARS-CoV-2, ribavirin for RSV, and sofosbuvir (Sovaldi) for HCV. Regulatory submission is targeted for 2027, with clinical trials estimated to commence in 2027.

MDL-4102 — a highly potent and selective BRD4 inhibitor with no measurable activity against BRD2 or BRD3. The program was optimized for BRD4 selectivity, transcriptional impact, and drug-like properties simultaneously. MDL-4102 is designed to overcome the dose-limiting hematologic toxicities that hindered prior pan-BET inhibitors, positioning it as a next-generation transcriptional therapy with the potential for durable efficacy across BRD4-driven malignancies. Beyond oncology, BRD4 biology extends into fibrosis, I&I, CVD, and neurodegenerative diseases, expanding the potential addressable market to more than $60 billion annually across these therapeutic areas. MDL-4102 is positioned as a differentiated, next-generation approach to targeting core disease-driving gene expression programs. Regulatory submission is targeted for 2027.

AmesNet™ is the first public release from Model Medicines’ multi-modal suite of ADMET GALILEO™ agents. To date, Model Medicines has only utilized these agents internally to discover its first-in-class therapeutics MDL-001 and MDL-4102. Together, the on/off target and ADMET agents are designed to enable Model Medicines to design, prioritize, and de-risk compounds better than any company in the field.

About Model Medicines

Model Medicines is an AI-first biotechnology company engineering first-in-class small molecules that target the biological choke points underlying disease. The company's drug franchise lifecycle is powered by GALILEO™, a proprietary end-to-end AI drug discovery and development platform. Through this platform, its core therapeutic programs target high-value mechanistic drivers of pathology, focusing specifically on viral replication via the novel allosteric RdRp Thumb-1 pocket and transcriptional control via the master regulator BRD4. These programs feed into a scalable "Pipeline-in-a-Pill" strategy, anchored by lead assets MDL-001 and MDL-4102 to address over 30 indications across virology, oncology, I&I, CVD, and neurodegenerative diseases. By leveraging large-scale computation to uncover entirely new classes of therapeutics once thought unreachable, Model Medicines is redefining what is possible in modern drug discovery. Learn more at www.modelmedicines.com. 

Media Contact

Patrick O’Neill

Head of Partnerships & Investor Relations

media@modelmedicines.com

www.modelmedicines.com