Further Comparative Studies of Fragment Weighting Schemes for Substructural Analysis

Abstract

This paper considers the use of fragment weighting schemes for substructural analysis studies. Experiments are reported using 2-D and 3-D substructural descriptors with both small and large datasets for which qualitative activity data are available. The results support previous work in suggesting the potential of probabilistic models of weighting first used in studies of text retrieval.to whom all correspondence should be addressed

Identification of Biological Activity Profiles Using Substructural Analysis and Genetic Algorithms

“…For example, the feature HBD is represented by 20 weights, HBDw 0 , HBDw 1 , ...HBDw 19 , which are the weights for exactly zero HBDs in a molecule, exactly one HBD, and up to >19 HBDs in a molecule, respectively. The weights are Ormerod et al 10,11 reported a detailed comparison of the many different weighting schemes that have been suggested for use in substructural analysis studies. They concluded that a weight that was originally used for text retrieval, referred to as R2, was the most effective.…”

“…There have been many reports of the use of substructural methods, using a range of different types of fragment weight. [6][7][8][9][10][11][12][13] Substructural analysis is normally applied to data from within an organization's corporate database, typically using the fragment substructures that form the basis for the initial, screening stage of a two-dimensional (2D) chemical substructure search and using activity data from primary biological screening results. However, there is no reason in principle why it cannot also be applied to data extracted from the many large databases of molecules that are now publicly available, using different types of (sub)structural feature and using nonspecific activity classes.…”

Evaluation of machine-learning methods for ligand-based virtual screening

“…A score is then computed for a test-set molecule by summing (or otherwise combining) the weights of those bits that are set in its fingerprint, this sum representing the overall probability of activity for that molecule given that it contains a particular pattern of bits. Substructural analysis was studied in considerable detail by workers at the National Institutes of Health in an extended programme to develop novel anti-cancer agents [47][48][49], and also by workers at Lederle [29] and Sheffield [50][51][52]. However, it is only in the last few years that this general approach has become widely used [53][54][55][56][57][58][59][60][61][62].…”

Mapping of Activity through Dichotomic Scores (MADS): A new chemoinformatic approach to detect activity‐rich structural regions

“…In literature there are several proposals to evaluate the similarity of compounds to existing known drugs by either structural features or chemical properties or a combination of structural and chemical properties. Many of them rely on the calculation of a scoring function that measures the overall probability of activity of a compound by summing (or otherwise combining) the weights for its constituent substructures …”

“…Often, substructure‐based screening relies on the calculation of weights that relate the presence in a molecule of a specific substructure moiety to the probability that the molecule is active in a predefined biological/toxicological test system. The probability of any new, untested molecule, in this way, can be simply determined additively, that is, by considering the activity‐related weights of each of its substructural fragments.…”