A novel approach for screening discrete variations in organic synthesis

Abstract

SUMMARYIn this paper we present a general strategy for screening discrete variations in organic synthesis. The strategy is based upon principal properties, i.e. principal component characterization of the constituents defining the reaction system. The first step is to select subsets of test items from each class of constituents defining the reaction space, i.e. substrates, reagents, solvents, catalysts, etc., so that the selected items from each class cover the properties considered. The second step is to construct a candidate matrix which contains all possible combinations of the items in the subsets. This matrix is a full multilevel factorial design. The third step is to assign a tentative model for the screening experiment and to construct the corresponding candidate model matrix. The fourth step is to select experiments to yield an experimental design that spans the variable space efficiently and that also gives good estimates of the model parameters. We present an algorithm that uses singular value decomposition to select experiments. The proposed strategy is then illustrated with an example of the Fischer indole synthesis.

Solvent subset selection for polymorph screening

“…Principal Properties selection is based on PCA and deflation of the data matrix X by orthogonal projection [8]. The first step is to perform PCA on the data matrix.…”

“…Commonly used approaches are Federov's D-optimality algorithm [6] and the Kennard-Stone selection method [7]. An intriguing contribution to this subject was made by Carlson et al [8], who demonstrated how the Principal Properties of a sample can be used as selection criteria for identifying optimal chemical agents in organic synthesis. In this work, we compare these three chemometric subset selection methods with a Cluster-Based approach as commonly used in data-mining [9,10].…”

Calibration Transfer Methods

“…Subset selection algorithms are also of interest outside calibration transfer ( Allesø et al , 2007 ). The most often used algorithms in the literature are the leverage-based methods, the Kennard-Stone selection ( Kennard and Stone, 1969 ) algorithm and principal properties selection ( Carlson et al , 2001 ). The Kennard-Stone algorithm adds samples that are farthest away from the previously selected set of points.…”

Principal Properties and Designs for Discrete Variations1

“…This can be accomplished by selecting new solvents for which the corresponding row in C is most parallel to the singular vector v i that corresponds to the smallest singular value σ i of X . A full account of these aspects is given in ref . By this principle, additional solvents can be selected to improve the precision of the estimated model parameters, and the following next six complementary solvents are: morpholine, pyridine, 1,2-dichloroethane, cis -decaline, 1,2-diaminoethane, and diglyme.…”

“… a The numbers within parentheses refer to the run number in the entire data set (162 runs) given in refs , . b Percent regioisomeric excess. …”