La Jolla, CA — [April 2, 2025] — Model Medicines, a pioneering biotech company leveraging artificial intelligence (AI) to accelerate drug discovery, announced today the publication of its latest preprint titled "AmesNet: A Deep Learning Model Enhancing Generalization in Ames Mutagenicity Prediction."  The preprint highlights AmesNet™, a groundbreaking adaptation of Model Medicines' proprietary ChemPrint™ molecular geometric convolutional neural network, designed specifically for predicting strain-specific Ames mutagenicity.

The Ames test, a cornerstone of preclinical safety assessment required for IND submissions, is costly and difficult to scale. AmesNet offers an AI-driven alternative capable of delivering real-time, strain-specific predictions of mutagenicity across eight bacterial strains used in traditional testing, effectively simulating the full Ames assay.

This breakthrough positions AmesNet™ as a superior alternative to existing computational models used for predicting mutagenicity, achieving state-of-the-art balanced accuracy and unprecedented sensitivity, particularly in predicting outcomes for chemically novel, out-of-domain compounds.

Key findings of the AmesNet paper include:

Enhanced Predictive Accuracy: AmesNet™ outperformed established QSAR methods, demonstrating balanced accuracy improvements of approximately 10% and sensitivity improvements of approximately 30%, marking a substantial advancement over previous best-in-class methodologies.

Superior Generalization: AmesNet™ notably improved prediction capabilities for out-of-domain chemical structures, an essential feature for robust mutagenicity assessment, reducing the risk of false negatives in preclinical screening.

Sensitivity in Critical Classes: AmesNet™ accurately identified mutagenicity in polyaromatic compounds—a traditionally challenging chemical class—demonstrating its ability to enhance early detection of high-risk genotoxic compounds.

Integration into GALILEO™ Platform: AmesNet™ is integrated within Model Medicines’ AI-driven GALILEO™ platform, further enriching a comprehensive ecosystem designed to streamline the drug discovery pipeline, reduce cost, and accelerate timelines to clinical entry.

Generative AI Drug Discovery

The current paper underscores the critical need for the deployment of innovative drug discovery platforms capable of engineering precise solutions for diseases with high-unmet medical need when traditional, dogmatic approaches fail. The GALILEOTM platform uniquely enables the generation of vast chemical libraries, while discovering libraries of novel, specific, and potent potential therapeutics. 

"Accurate mutagenicity prediction is vital to drug discovery, yet existing methods frequently suffer from critical blind spots," said Dr. Daniel Haders, CEO of Model Medicines and corresponding author. "With AmesNet™, we have demonstrated significant improvements in sensitivity and generalization, addressing unmet needs in preclinical safety assessment. This advancement further underscores the potential of AI-driven methodologies to transform drug discovery from initial target identification to final clinical validation."

Model Medicines’ GALILEO™ platform continues to set new benchmarks in integrating AI across drug discovery and development processes. The addition of AmesNet™ significantly enhances GALILEO’s ability to ensure drug candidates are both potent and safe, accelerating the pathway from discovery to clinical application.

AmesNet™ strengthens Model Medicines' position at the forefront of regulatory-grade predictive modeling. While not intended to replace the physical Ames test in current FDA workflows, AmesNet™ dramatically enhances early-stage screening by improving the identification of true positive Ames compounds – ensuring only the safest, most viable compounds progress to expensive lab testing.

The platform also opens up licensing and partnership opportunities for CROs and biopharma companies seeking to improve the preclinical safety profiling of their discovery libraries through AI. Model Medicines plans to expand GALILEO™ with additional models targeting absorption, metabolism, and systemic toxicity in the near future.

Additional Preprints

GALILEO Generatively Expands Chemical Space and Achieves One-Shot Identification of a Library of Novel, Specific, Next Generation Broad-Spectrum Antiviral Compounds at High Hit Rates

MDL-001: An Oral, Safe, and Well-Tolerated Broad-Spectrum Inhibitor of Viral Polymerases

Discovery of RdRp Thumb-1 as a novel broad-spectrum antiviral family of targets and MDL-001 as a potent broad-spectrum inhibitor thereof

ChemPrint: An AI-Driven Framework for Enhanced Drug Discovery