Dimensionality reduction techniques for mapping diphosphine chemical space using the Ligand Knowledge Base.

Identifier:  TFG:6569

Handle: https://hdl.handle.net/20.500.11797/TFG6569

Authors:  Villares Cañón, Mario

Abstract:
The chemical space is the multidimensional region where all known and unknown molecules are. Describing the diphosphine chemical space by the usage of the Ligand Knowledge Base (LKB) methodology, DFT-calculated property descriptors and the latter application of dimensionality reduction techniques lead to maps of chemical space that are useful for organometallic catalysts optimisation. Different dimensionality reduction techniques are tested (PCA, UMAP and t-SNE) and the information that such maps contain is determined by clustering algorithms. Structural characteristics have been used to see if the maps show trends. If structurally similar diphosphines are clustered together, this can likely be translated to similar catalytic performance since they will have similar properties. Test for comparing the cluster ability, robustness and extent of retained information are performed. At the end an experimental-based test case is carried out to demonstrate the potential that those generated maps and prediction models have on the task of optimising organometallic catalysts. The whole project was developed at the University of Bristol in Dr. Natalie Fey Group.