AmesNet: A Deep Learning Model Enhancing Generalization in Ames Mutagenicity Prediction
Revolutionizing Preclinical Drug Safety Assessment

At Model Medicines, we're pushing the boundaries of artificial intelligence in drug discovery, introducing AmesNet—a breakthrough deep learning model that transforms how we predict drug mutagenicity.
Key Achievements
Unprecedented Sensitivity: AmesNet achieved a sensitivity of 0.75, representing a 30% improvement over the best existing machine learning methods.
Enhanced Generalization: Unlike previous models, AmesNet demonstrates robust performance across both in-domain and out-of-domain datasets.
Advanced AI Integration: Built upon ChemPrint, our molecular geometric convolutional neural network with proven success in drug discovery programs.
From ChemPrint to AmesNet: A Strategic Evolution
AmesNet builds on ChemPrint's remarkable track record, which has already demonstrated impressive capabilities:
46% zero-shot hit rate in AXL and BRD4 oncology programs
100% one-shot hit rate in RdRp Thumb-1 antiviral program
Off-target binding reductions ranging from 800-fold to over 15,000-fold
Technical Innovation
AmesNet innovates by combining ChemPrint's molecular embeddings with specialized Ames mutagenicity features. This approach enables unprecedented predictive power in assessing potential drug candidates' genetic safety.
Future of Drug Discovery
AmesNet represents more than a single tool—it's part of our broader GALILEO ecosystem. We are systematically developing AI tools to accelerate and de-risk the drug discovery pipeline by comprehensively predicting ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) characteristics.
Read the Full Preprint
Dive deeper into the technical details and groundbreaking results of AmesNet. Full preprint available now.
Details
Date
Mar 20, 2025
Category
Pre-Print
Reading
2 Mins
Author

Patrick ONeill
Investor Relations
Related News