ZAO - 4D Molecular Foundation Model

ZAO is a 4D molecular foundation model developed by SyntheticGestalt for drug discovery and molecular property prediction. Unlike conventional models that represent molecules as 1D sequences or 2D graphs, ZAO processes multiple 3D conformations of each molecule simultaneously, capturing spatial arrangements, molecular flexibility, and charge distributions that flat representations miss.

Send SMILES strings via REST API and receive 2048-dimensional molecular embeddings. These embeddings serve as drop-in features for any downstream ML model, including CatBoost, XGBoost, scikit-learn, and neural networks. All preprocessing, including conformer generation and feature extraction, is handled inside the container with no external tools required.

ZAO embeddings fed into a simple CatBoost model achieve state-of-the-art performance across standard drug discovery benchmarks. On the TDC ADMET Benchmark (22 datasets), ZAO ranks #1 on the leaderboard in 8 datasets, including clearance, solubility, lipophilicity, CYP metabolism, and toxicity prediction, and top 3 in 9 datasets. On TDC DTI BindingDB_Patent, ZAO achieves PCC 0.687 (+0.1 over the previous leaderboard #1)

Processing multiple conformers per molecule is critical for accuracy. Ablation studies show that using 10 conformers improves activity prediction R2 by +12% and property prediction R2 by +4.8% compared to a single conformer, confirming that conformational flexibility information is essential for molecular understanding.

ZAO is designed for drug-like compounds with molecular weight between 100 and 1000 Da. Molecules outside this range return null embeddings.

Volume discounts are available through AWS Marketplace Private Offers. For large-scale or enterprise usage, please contact zao@syntheticgestalt.com  to request a Private Offer with custom pricing and terms.