Kipu Quantum uses QuEra's Aquila to help predict molecule toxicity

[Beth]

Kipu Quantum has released a technical report showing how a neutral atom quantum processor can help classify whether molecules are toxic. Quantum Computing Report says the experiments ran on QuEra's 256 qubit Aquila machine. The authors are Antonio Ferrer Sanchez, Carlos Flores and Alejandro Gomez Cadavid

The approach is a form of quantum feature extraction. For each molecule in the UCI Molecular Toxicity dataset, 200 atoms arranged on a 14 by 15 grid were driven with pulses lasting under four microseconds. Measuring the atoms over 500 shots produced around 1,000 quantum features per molecule

Those features were then fed into ordinary machine learning models, including XGBoost, CatBoost, Random Forest, support vector machines and Google's TabFM. The team tested them across 100 different data splits to make sure the results were not a fluke of one particular split

The improvements look meaningful. Balanced accuracy rose from about 0.55 to 0.70, the F1 score roughly doubled, and the best area under the curve was about 0.81 using a support vector machine. In other words, the quantum features seemed to help the classical models pick out toxic molecules more reliably

These are the company's own results and have not been independently verified, and the original blog post dates from July, so this is a write up of earlier work. Still, using quantum hardware to generate features for classical models is a practical idea that suits today's noisy machines. It does not need a fully fault tolerant computer

Can quantum feature extraction beat strong classical baselines? And is drug safety a realistic early use for quantum computing?

Kipu Quantum Demonstrates Neutral-Atom Quantum Feature Extraction for Molecular Toxicity Classification on QuEra Hardware - Quantum Computing Reportquantumcomputingreport.com