Combinatorial Drug Discovery from Activity-Related Substructure Identification

A newly developed drug discovery method composed of graph theoretical approaches for generating structures combinatorially from an activity-related root vertex, prediction of activity using topological distance-based vertex index and a rule-based algorithm and prioritization of putative active compounds using a newly defined Molecular Priority Score (MPS) has been described in this chapter. The rule-based method is also used for identifying suitable activity-related vertices (atoms) present in the active compounds of a data set, and identified vertex is used for combinatorial generation of structures. An algorithm has also been described for identifying suitable training set–test set splits (combinations) for a given data set since getting a suitable training set is of utmost importance for getting acceptable activity prediction. The method has also been used, to our knowledge for the first time, for matching and searching rooted trees and sub-trees in the compounds of a data set to discover novel drug candidates. The performance of different modules of the proposed method has been investigated by considering two different series of bioactive compounds: (1) convulsant and anticonvulsant barbiturates and (2) nucleoside analogues with their activities against HIV and a data set of 3779 potential antitubercular compounds. While activity prediction, compound prioritization and structure generation studies have been carried out for barbiturates and nucleoside analogues, activity-related tree–sub-tree searching in the said data set has been carried for screening potential antitubercular compounds. All the results show a high level of success rate. The possible relation of this work with scaffold hopping and inverse quantitative structure–activity relationship (iQSAR) problem has also been discussed. This newly developed method seems to hold promise for discovering novel therapeutic candidates.