Datalab Seminar

Heat of formation of azabole derivatives: Can Machine Learning be applied?

Speaker:  Maxime Ferrer, Ibon Alkorta, Jose Elguero (Instituto de Química Médica (CSIC))
Date:  Friday, 09 June 2023 - 12:00
Place:  Aula Gris 2, ICMAT


The azabole derivatives are molecules composed of three joined rings, which contain two boron atoms in the central ring, as well as nitrogen and carbon atoms in the remaining positions (positions 1 to 10) [1]. Azabole chemistry has been known for a long time. These compounds are stable, and it is easy to modify the substituents on the boron atoms, allowing for the generation of new azaboles. Currently, azaboles are utilized in coordination chemistry as potential ligands [2].

The heat of formation of a molecule is a physical quantity that represents the energy required to form the molecule. It can be understood as the energy needed to bring the atoms of the molecule together and bond them. This energy, also known as the enthalpy of formation, provides a measure of the stability of a given system. Generally, the larger the enthalpy of formation, the less stable the molecule.

Theoretical chemists have developed various techniques to compute the heat of formation of any system beforehand. However, these methods still require significant computational resources, as they involve the combination of several high-level ab initio methods to obtain results that closely match experimental data.

In our study, we calculated the enthalpy of formation for over 700 azabole derivatives. These derivatives were obtained by keeping the boron atoms fixed and generating all possible structures with n atoms of nitrogen (where 3 < n < 10) and (10 - n) atoms of carbon in the remaining positions.

Our objective is to use the formalisms of Machine Learning to develop a model that takes the relative position and type of atoms as input and predicts the heat of formation of the molecule as output.

[1] J. Am. Chem. Soc., 89 (1967), 4948
[2] Coordination Chemistry Reviews. 473 (2022), 214812