Written by Ju-Won Park 

Quantum chemistry is the study of the properties of molecules and reactions. Recently,  computer technologies have started to be used to understand, model, and predict molecular properties and reactions. This consists of the properties of nanometer materials (substances with dimensions smaller than 100 nanometers) and processes/reactions of biological systems.

There is a lack of electronic degrees of freedom in molecules that are key to quantum chemistry. It is important to know that all chemical concepts and physical molecular properties can be accurately modeled with the electronic Schrodinger equation and come from the ground-state (lowest energy state) wavefunction. The wavefunction is all measurable information about a particle. In the study, a team of researchers led by the University of Warwick developed a deep machine learning algorithm that can predict the wavefunctions of molecules, which determine all properties of molecules; this could then be used to predict the outcomes of experiments. The algorithm will be able to construct predictive models that can find new sustainable catalyst materials and design new synthetic pathways for drug delivery. 

The model provides access to electronic properties that are important for chemical interpretations of reactions such as charge populations, bond orders, as well as dipole and quadrupole moments without the need for a specialized ML model for each property. 

But what is deep machine learning? Grossly summarized, neural networks are modeled on the human brain. It consists of millions of nodes that are interconnected. Nodes are intersections or connecting points. Most neural nets have multiple layers of nodes and data moves through them in one direction (this is called “feed-forward”). When training data is fed to the input layer, it passes through multiple layers which changes the weight and thresholds so that training data with the same labels have similar outputs. Data is hand-labeled to help object recognition systems find visual patterns in images.

Originally, the equations solved by the AI required months of computing time but this algorithm can give accurate predictions within seconds on a laptop or phone. . The algorithm predicts wave functions based on quantum chemistry data. This data can be used in other quantum chemical calculations.

This work serves as an example for future applications of both electronic and structural properties of a molecule in the discovery process in both computational chemistry and molecular physics. Some find applications in medicine for faster drug discovery, which could reduce R&D resources needed for pharmaceuticals.

Works Cited

Artificial Intelligence algorithm can learn the laws of quantum mechanics and speed up drug delivery. (2019, November 20). Retrieved from https://www.worldpharmanews.com/research/5025-artificial-intelligence-algorithm-can-learn-the-laws-of-quantum-mechanics-and-speed-up-drug-delivery.

Quantum Chemistry. (n.d.). Retrieved December 11, 2019, from https://www.sciencedirect.com/topics/chemistry/quantum-chemistry.

Schütt, K. T., Gastegger, M., Tkatchenko, A., Müller, K.-R., & Maurer, R. J. (2019). Unifying machine learning and quantum chemistry with a deep neural network for molecular wavefunctions. Nature Communications, 10(1). doi: 10.1038/s41467-019-12875-2

Wavefunction Properties. (n.d.). Retrieved December 11, 2019, from http://hyperphysics.phy-astr.gsu.edu/hbase/quantum/wvfun.html#c2.