Decoding Chemical Processes: The Power of Data-Driven Descriptors

Chemical descriptors are mathematical constructs used to extract insights from chemistry. Over the past decade,there has been an exponential growth in their applicability. This has transformed chemical featurization into an active area of research. In this thesis, building on previous work from our group, we sought optimal chemical representations to uncover the key forces governing diverse chemical scenarios. In this context, we first tested the applicability of an alternative mathematical object namely the hidden descriptors to envisage the electronic tendencies of N-heterocyclic carbene ligands. Additionally, employing the same hidden descriptor tool, we attempted to pinpoint suitable metal fragments for coordinating monohapto hydrogen ligands. We further delved into the capabilities of the hidden descriptor approach to identify kinetic forces of an organic bimolecular nucleophilic substitution reaction. In this transformation, the role of the nucleophiles was described in a mathematical optimal manner. On the other hand, we also developed a different methodology, namely AABBA, designed to derive fixed-length molecular representations of complexes from molecular graphs. We assessed the effectiveness of this novelvectors within two regression tasks.

Therefore, this thesis provides a deep theoretical knowledge about the alternative chemical descriptors that efficiently describe and correlate target properties.