AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-19
Spatial statistics for screening molecular structures
Computational materials discovery typically relies on deep architectures that need millions of DFT-labeled structures, which is impractical for data-scarce molecular screening. This review proposes spatial-statistics feature engineering: molecules are voxelized into scalar fields, two-point correlations are computed via FFT, and PCA yields low-dimensional convex representations for lean networks and surrogate models. The approach reaches sub-2% prediction error with as few as 10 training samples across periodic crystals, disordered high-entropy alloys, and organic molecules.