Field-based similarity search system and method

摘要

A field-based similarity search system includes an input device which inputs a query molecule, and a processor which partitions a conformational space of the query molecule into a fragment graph including an acyclic graph including plural fragment nodes connected by rotatable bond edges, computes a property field on fragment pairs of fragments of the query molecule from the fragment graph, the property field including a local approximation of a property field of the query molecule, constructs a set of features of the fragment pairs based on the property field, the features including a set of local, rotationally invariant, and moment-based descriptors generated from all conformations of the fragment graph of the query molecule, and weights the descriptors according to importance as perceived from a training set of descriptors to generate a context-adapted descriptor-to-key mapping which maps the set of descriptors to a set of feature keys.

主权项

1. A field-based similarity search system, comprising:
an input device which inputs a query molecule; and a processor which:
partitions a conformational space of said query molecule into a fragment graph comprising an acyclic graph including plural fragment nodes connected by rotatable bond edges;computes a property field on fragment pairs of fragments of said query molecule from said fragment graph, said property field comprising a local approximation of a property field of said query molecule;constructs a set of fragment pairs for description of each candidate molecule wherein each set is known as a set of candidate molecule fragment pairs;constructs a set of features of said fragment pairs based on said property field, said features comprising a set of local, rotationally invariant, and moment-based descriptors generated from all conformations of said fragment graph of said query molecule; andweights said descriptors according to importance as perceived from a training set of descriptors to generate a context-adapted descriptor-to-key mapping which maps said set of descriptors to a set of feature keys comprising indices that label grid cells in discriminant space, wherein a candidate molecule fragment pair is within a set of candidate molecule fragment pairs that describe a candidate molecule stored in a feature database to said query molecule, and wherein the number of candidate molecule fragment pairs in the set, Cfp, is given by Cfp=(n−1)×(Cfrag)2×Crbe, where n is the number of fragments in said candidate molecule and n−1 is the number of rotatable bond edges connecting n fragments, Cfrag is a number of conformations in a fragment of said fragments and Crbe is a number of steps in which said rotatable bond edges are sampled.