Method for accurately estimating pKa of molecules using atom type definitions and partial least squares
View Patent ↗A computer implemented software method enables the prediction of the pK a of an arbitrary molecule based upon a knowledge of the molecular structure of that molecule and a statistical analysis of the molecular structures of a group of molecules (training set) for which the pK a is known. Hierarchical atom connectivity trees are constructed for the training set and the various atoms types identified in each molecule are associated in a bit string for that molecule and also associated with the experimentally determined pK a for that molecule. PLS analysis of the training set data yields coefficients associated with each atom type represented in the bit strings. A hierarchical atom connectivity tree may then be constructed for the molecule of interest. The predicted pK a is determined by multiplying the number of occurrences of each atom type in the molecule of interest by the PLS coefficient determined for that atom type and summing the resulting multiplications.
1. A software implemented methodology for determining the pK a of a molecule of interest comprising the following steps:
a) determine the hierarchical atom type connectivity tree noting the number of times each atom type occurs for each ionizable group on each molecule of a series of molecules for which an experimentally determined pK a is known;
b) place each of the determined atom types noting the number of times each atom type occurs for each hierarchically determined connectivity tree in a row of a data table along with the experimentally determined pK a of the molecule from which the tree was determined;
c) using the partial least squares (PLS) statistical methodology, extract coefficients associated with each atom type represented at each hierarchical level;
d) determine the hierarchical atom type connectivity tree noting the number of times each atom type occurs for the molecule of interest; and
e) multiply the number of occurrences of each atom type in the molecule of interest by the PLS coefficient determined for that atom type and sum the resulting multiplications to obtain the predicted pK a .
2. The method of claim 1 in which each atom type from the hierarchical atom connectivity tree in steps a and d for each ionizable group on each molecule is placed into a separate bin in a bit string.
3. The method of claim 2 in which the extracted coefficients are associated with the appropriate bin in the bit string.