Using a data-driven approach, the researchers discovered "design freedom" in molecular structures that results from weak correlations in quantum mechanical properties. This discovery, combined with machine learning, could revolutionize molecular design and drug discovery.
Graphical description of the rational molecular design process, which involves a "needle in a haystack" search for molecules with desired properties. Image credit: Leonardo Medrano Sandonas, University of Luxembourg; background image provided by rawpixel.com on Freepik
The use of data-driven methods to explore the vast space of molecules and materials has inspired countless academic and industrial communities to search for the fundamental relationships that exist between molecular structural features and their physicochemical properties. Although significant progress has been made in this field, a comprehensive understanding of these complex relationships is still lacking, even in the more manageable field of chemistry and chemical substances, small molecules, despite the critical importance and high relevance of these molecules throughout the chemical and pharmaceutical sciences.
Alexandre Tkatchenko, Professor of Theoretical Chemical Physics at the Department of Physics and Materials Science at the University of Luxembourg, said: "Revealing the complex relationship between molecular structure and properties will not only provide us with the tools needed to explore and characterize molecular space, but also greatly improve our ability to rationally design molecules with a range of targeted physicochemical properties."
Weak correlation brings "design freedom"
In a paper titled "'Design freedom' in the space of chemical compounds: Towards Rational in Silico Design of Molecules with Targeted Quantum-Mechanical Properties" published in the prestigious journal Chemical Science, an important finding is that most quantum mechanical properties of small molecules are only weakly correlated.
Robert Di Stasio Jr., a professor of theoretical chemistry at Cornell University, said: "While one might initially view this discovery as a challenge to rational molecular design, our analysis highlights the inherent flexibility of CCS, in which there appear to be few constraints that prevent a molecule from exhibiting any pair of properties at the same time, or that prevent many molecules from sharing a range of properties."
Finding the best path in chemical space
The molecular design process usually involves the simultaneous optimization of multiple physicochemical properties. To explore how this inherent flexibility will be reflected in the molecular design process, the authors used Pareto multi-property optimization method to find molecules with both macromolecule polarity and electronic gap. This is a design task related to identifying new molecules for polymer batteries. The authors found several unexpected pathways through chemical space connecting molecules through structural and/or compositional changes, reflecting the freedom to rationally design and discover molecules with targeted property values.
Professor Tkatchenko explains: "A potentially interesting next step would be to combine these Pareto-optimal structures with powerful machine learning methods to establish a reliable multi-objective framework for systematic navigation of hitherto unexplored chemical space."
Impact on molecular design paradigms
"By demonstrating that 'design freedom' is a fundamental and emerging property of CCS, our work has a series of important implications for the fields of rational molecular design and computational drug discovery. First, we hope that this work will challenge the chemical science community to consider how this inherent flexibility can be exploited to extend the dominant paradigm in forward molecular design processes." Theoretical Chemistry, University of Luxembourg "We also hope that this work will make substantial progress in solving inverse molecular design problems, in which one tries to find a molecule (or a set of molecules) that corresponds to a set of target properties," explains Dr. Leonardo Medrano Sandonas, a postdoctoral researcher in the Physics Group.
Combining insights gained from this work with advanced machine learning methods will help develop effective strategies for high-throughput screening of novel molecules for specific applications, a prominent research direction of Professor Tekatchenko's research group.