Impact of Phosphine Featurization Methods in Process Development

Modern
high-throughput experimentation (HTE) has enabled the rapid
exploration of large expanses of chemical reaction space to accelerate
the development of key synthetic steps in pharmaceutical processes.
However, the dimensionality of reaction parameters, the desire to
use minimal starting material, and the need to thoroughly analyze
reaction outcomes still require the judicious selection of which experiments
to perform in which order. Therefore, the development of a capability
to quantify reagent diversity and analyze reaction outcomes in HTE
holistically is paramount. A method to address this goal would combine
computational featurization of key reaction components with the use
of multivariate linear regression modeling to correlate the reaction
performance outputs. In this context, we describe a process of establishing
a computational featurization platform for monodentate phosphine ligands
and considerations for its implementation at GSK. We demonstrate that
the choice of computational method has an impact on phosphine descriptor
values, ligand selection for experiments, and the development of linear
regression models of reaction outcomes.