AQCat - AI Models for Catalyst Discovery

AQCat is a family of spin-aware machine learning interatomic potentials (MLIPs) built for heterogeneous catalysis screening and chemical discovery. Deployed as an Amazon SageMaker model package in the customer own AWS tenant, AQCat supports both direct model inference for structural energies and forces and an adsorption workflow for estimating minimum adsorption energies across candidate catalyst surfaces. It is designed to deliver DFT-level adsorption energy accuracy at up to 20,000x the speed of physics-based calculations. Developed for materials scientists, catalyst researchers, computational chemists, and R&D teams in industries including chemicals, energy, oil and gas, agriculture, and pharmaceuticals, it helps users evaluate catalyst candidates faster while keeping proprietary structures and workflows inside their own environment.

The AQCat25-EV2 model is trained on the AQCat25 dataset with 13.5 million high-fidelity DFT calculations. In published benchmarking referenced in the listing draft, it identified the global minimum adsorption within 0.1 eV for 69.9% of the test set and achieved mean absolute errors of 349 meV on energies and 16.98 meV/A on forces on the AQCat25 test set, while maintaining 290 meV energy error and 20.46 meV/A force error on a subsampled OC20 validation set. For organizations seeking a machine learning force field (MLFF) for adsorption energy prediction, transition-state exploration, and virtual screening, AQCat is designed to reduce the tradeoff between quantum-chemistry accuracy and computational throughput.

Revisions to Standard Contract for AWS Marketplace: We are making our model available under the Standard Contract for AWS Marketplace, but note that our professional liability insurance does not cover proprietary rights infringement as described in Section 11.1.2 of the Standard Contract, and this product listing amends that section of the contract as permitted by Section 1.3 of the Standard Contract.