AI Tool Predicts Stable Molecular Forms to Speed Drug Discovery

New York University researchers developed an AI model called Tautomer-Predictor to identify the most stable tautomer—the form of a molecule defined by the position of its hydrogen atoms—from a 2D chemical structure.

The graph neural network learned stability patterns from crystallographic data and can predict stable tautomers without requiring 3D structures or expensive quantum-mechanical calculations.

The model was tested on crystal-structure, aqueous-solution, and drug-like benchmarks, where researchers reported strong and transferable performance.

The team says the open-source tool can screen millions of compounds rapidly; it reportedly analyzed about 4.6 million drug-like molecules in hours.

The advance could improve virtual screening and molecular design by reducing errors caused when different tautomeric forms have different biological activity, binding behavior, solubility, or safety profiles.

The method is a computational aid rather than a replacement for laboratory confirmation, since molecular stability can depend on conditions such as solvent, pH, temperature, and crystal environment.