Chem World: A Large-Scale Benchmark and Physics-Informed Framework for Trustworthy Chemical Property Prediction

The paper introduces Chem World, a unified benchmark that combines 17 chemical datasets and more than 800,000 molecular samples for property-prediction tasks including density, electrical conductivity, and solubility. It also proposes Mixture-PINN, a physics-informed neural-network framework intended to incorporate chemical prior knowledge into data-driven prediction. The abstract reports improved accuracy, robustness, and reliability over existing methods, but provides no numerical results, dataset-by-dataset breakdown, model configuration, or detailed description of the physical constraints.