Revolutionary physics-based machine learning

Our core technologies use physics-based machine learning features, derived from efficient quantum mechanical calculations, to map chemical space with unprecedented fidelity and transferability.

Learn More
OrbNet workflow
machine learning

Machine learning

We use features derived from mean-field-cost quantum calculations, either in a molecular orbital or atomic orbital representation. The resulting ML models have unprecedented transferability, and our deep-learning architectures can be trained for both energy and property predictions. ML for energy predictions provides thousandfold improvements in efficiency relative to conventional DFT methods. Quantum-mechanical features compactly represent chemical space for the efficient machine-learning of high-value molecular properties.


Qcore Simulation Engine

Our Qcore quantum simulation engine integrates state-of-the art implementations of standard methods from quantum chemistry with machine-learning featurization and prediction engines and a suite of unique features for quantum embedding.

Core methods are available such as Hartree-Fock theory, density functional theory, and modern semiempirical methods, all wrapped in an efficiently parallelized application, developed by a team of dedicated software engineers. Tools for chemistry (such as constrained and unconstrained geometry optimization, transition-state search, ab initio dynamics, frequencies, etc) are available for all methods.

Install with conda and use our Qcore tutorials to get started.