Functional groups as a language for property prediction

Mark Heezen, Manuel Alcamí, Clemens Rauer, Freija De Vleeschouwer

Quantitative Structure Property Relationships (QSPR) are powerful tools to correlate molecular structures with their properties and have been used extensively in the last decades [1,2]. We aim to use a QSPR approach to predict thermodynamic molecular properties by using molecular functional groups as structural features. This will lead to new chemical insights which will be useful for molecular design.

To provide an explainable relationship between molecular structures and their physicochemical properties, we choose to use conventional functional groups as implemented in the Checkmol software [3] over generating substructures with unsupervised Machine Learning algorithms as proposed in [4].

For our method development, we focus on an extensive database of pesticide structures, developed within the SEPIA-pesticides project (sepia-pesticides.es). This database includes a comprehensive conformational analysis and a full optimisation of the most stable conformers obtained using the same computational protocol. Utilising this database, we aim to develop robust supervised learning models for predicting the vapour pressure.

Looking ahead, we envision broader applications of this methodology in inverse molecular design, enabling the tailored synthesis of molecules with desired physicochemical properties [5].

1. Roy, K., Kar, S., and Das R.N., A Primer on QSAR/QSPR Modeling 2015, 1-36.
2. Yousefinejad, S., and Hemmateenejad, B., J. Chemolab 2015, 149(B), 177-204.
3. Haider, N., Molecules 2010, 15(8), 5079-5092.
4. Li, B., Lin, M., Chen, T., and Wang, L., Brief. Bioinform. 2023, 24(6), 398.
5. Gantzer, P., Creton, B., and Nieto-Draghi, C., Mol. Inf. 2020, 39, 1900087.

 Mark Heezen