Mining molecular fragments: finding relevant substructures of molecules

摘要: We present an algorithm to find fragments in a set of molecules that help discriminate between different classes for instance, activity drug discovery context. Instead carrying out brute-force search, our method generates by embedding them all appropriate parallel and prunes the search tree based on local order atoms bonds, which results substantially faster eliminating need frequent, computationally expensive reembeddings suppressing redundant search. prove usefulness demonstrating activity-related groups chemical compounds well-known National Cancer Institute's HIV-screening dataset.