Machine Learning Breakthrough Enables Precise Quantum Chemistry Calculations for Drug Discovery

Scientists have achieved a major breakthrough in computational chemistry by applying machine learning to solve a decades-old dilemma in quantum chemistry, enabling precise and stable calculation of molecular energies and electron densities with significantly less computational power. The new orbital-free approach permits calculations for very large molecules, including "drug-like" molecules, and demonstrates stable convergence for the first time using this method.

The breakthrough involves a new process called STRUCTURES25, which is based on a specifically developed neural network that learns the relationship between electron density and energy directly from precise reference calculations. The model captures the chemical environment of each individual atom in a mathematically detailed representation. A unique training concept was pivotal: the model was trained not only with converged electron densities but also with many variants surrounding the correct solution – generated by targeted, controlled changes in the underlying reference calculations.

This computing process is therefore able to reliably find a physically meaningful solution for molecular energies and electron densities even in case of small deviations. It remains stable without "getting lost" in the calculation. In tests on a large and diverse collection of organic molecules, STRUCTURES25 achieved a precision that can compete with established reference calculations. Initial runtime comparisons prove that the computing process can scale better with growing molecule size and hence increase the speed of the calculation.

How electrons are distributed in a molecule determines its chemical properties – from its stability and reactivity to its biological effect. Reliably calculating this electron distribution and the resulting energy is one of the central functions of quantum chemistry. These calculations form the basis of many applications in which molecules must be specifically understood and designed, such as for new drugs, better batteries, materials for energy conversion or more efficient catalysts.

In quantum chemistry, molecules are frequently described using density functional theory, which allows for the fundamental prediction of chemical molecular properties without having to calculate the quantum mechanical wave function. The electron density is used as the main quantity instead – a simplification that finally makes computations practicable. This orbital-free approach promises especially efficient calculations but until now was considered barely useful, since small deviations in the electron density led to unstable or "non-physical" results.

Separately, researchers have also made progress in quantum computing approaches to molecular simulations. Ground-state energy estimation for molecular systems with active spaces of 20-50 spatial orbitals is now achievable using approximately 100,000 physical qubits. Classical full configuration interaction methods, considered the 'gold standard' for molecular electronic structure calculations, become computationally insufficient for modelling systems beyond approximately 20 electrons in 20 orbitals, a limitation known as the 'exponential wall'.

This restricts accurate simulations of complex molecules crucial for understanding chemical reactions, material properties, and biological processes. The team demonstrated broad applicability beyond simple molecules by applying their framework to models of iron-sulfur clusters, which play vital roles in biological electron transfer, cytochrome P450 active sites, important in drug metabolism and detoxification, and catalysts designed for carbon dioxide utilisation.