Simple algebraic operations on hyper dimensional fingerprints can predict chemical properties faster than massive neural networks.
Molecular discovery usually requires training resource-heavy Graph Neural Networks on thousands of examples. These hyper-dimensional fingerprints provide a deterministic alternative that requires zero training time. This method out-performs conventional models while running on a fraction of the hardware. It allows researchers to screen vast chemical libraries for new drugs or materials at a negligible cost. This shift makes high-fidelity molecular prediction accessible to every chemist with a standard laptop. We can now discover new molecules using simple math instead of expensive supercomputers.
Hyper-Dimensional Fingerprints as Molecular Representations
arXiv · 2604.27810
Computational molecular representations underpin virtual screening, property prediction, and materials discovery. Conventional fingerprints are efficient and deterministic but lose structural information through hash-based compression, particularly at low dimensionalities. Learned representations from graph neural networks recover this expressiveness but require task-specific training and substantial computational resources. Here we introduce hyperdimensional fingerprints (HDF), which replace th