Title
|
|
|
|
Machine learning for vibrational circular dichroism : constructing novel and accelerating established applications
| |
Author
|
|
|
|
| |
Abstract
|
|
|
|
Determination of the so-called Absolute Configuration (AC) of a chiral compound is an important analytical step in research areas such as medicine and agrochemistry. The AC of a chiral compound can be determined with Vibrational Circular Dichroism (VCD), which measures the difference in absorbance of left and right circularly polarised infrared radiation. VCD spectra exhibit substantial chiral sensitivity and contain an abundance of conformational information. Unfortunately, there are no general empirical rules capable of linking a VCD spectrum to a specific AC or predicting the influence of the conformations on the VCD spectrum. Therefore, in each VCD application one has to rely on expensive (DFT) calculations for the conformers of all possible AC's. In this thesis, the added value of Machine Learning (ML) is explored for the AC determination workflow with VCD. The presented results demonstrate that ML models are capable of directly extracting the AC from the VCD spectrum after training these models on a set of structurally similar compounds. Neural Networks can successfully predict the influence of conformations on the spectrum from their geometries, as long as these conformations correspond to the same AC. Additionally, the potential of linear ML models is tested to determine the composition of terpene mixtures and natural oils. |
| |
Language
|
|
|
|
English
| |
Publication
|
|
|
|
Antwerpen
:
Universiteit Antwerpen & Universiteit Gent
,
2023
| |
Volume/pages
|
|
|
|
224 p.
| |
Note
|
|
|
|
:
Herrebout, Wouter [Supervisor]
:
Bultinck, Patrick [Supervisor]
| |
Full text (open access)
|
|
|
|
| |
|