Analysis of selection methodologies for combinatorial library design

We have implemented and adapted in Pralins (Program for Rational Analysis of Libraries in silico), the most popular sparse (cherry picking) and full array (sublibrary) selection algorithms: hierarchical clustering, k-means clustering, Optimum Binning, Jarvis Patrick, Pral-SE (partitioning techniques) and MaxSum, MaxMin, MaxMin averaged, DN2, CTD (distance-based methods). We have validated the program with an already synthesized three-component combinatorial library of FXR partial agonists characterized by standard computational chemistry descriptors as case study. This has let us analyze the goodness of both the partitioning techniques for space division and all the selection methodologies with respect to representativity in terms of population and space coverage for different selection sizes. Within the chemical space analyzed, both hierarchical clustering and Optimum Binning division strategies are found to be the most advantageous reference space divisions to be used in the subsequent population and space coverage studies. Complete hierarchical clustering appears also to be the preferred selection methodology for both sparse and full array problems. The full array restriction fulfillment can easily be overcome by convenient optimization algorithms that allow optimal reagent selection preserving > 90% of the population coverage.