Batista, José: Analysis of Random Fragment Profiles for the Detection of Structure-Activity Relationships. - Bonn, 2008. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5N-15839
@phdthesis{handle:20.500.11811/3701,
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5N-15839,
author = {{José Batista}},
title = {Analysis of Random Fragment Profiles for the Detection of Structure-Activity Relationships},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2008,
note = {Substructure- or fragment-type descriptors are effective and widely used tools for chemical similarity searching and other applications in chemoinformatics and computer-aided drug discovery. Therefore, a large number of well-defined computational fragmentation schemes has been devised including hierarchical fragmentation of molecules for the analysis of core structures in drugs or retrosynthetic fragmentation of compounds for de novo ligand design. Furthermore, the generation of dictionaries of structural key-type descriptors that are important tools in pharmaceutical research involves knowledge-based fragment design. Currently more than 5 000 standard descriptors are available for the representation of molecular structures, and therefore the selection of suitable combinations of descriptors for specific chemoinformatic applications is a crucial task.
This thesis departs from well-defined substructure design approaches. Randomly generated fragment populations are generated and mined for substructures associated with different compound classes. A novel method termed MolBlaster is introduced for the evaluation of molecular similarity relationships on the basis of randomly generated fragment populations. Fragment profiles of molecules are generated by random deletion of bonds in connectivity tables and quantitatively compared using entropy-based metrics. In test calculations, MolBlaster accurately reproduced a structural key-based similarity ranking of druglike molecules. To adapt the generation and comparison of random fragment populations for largescale compound screening, different fragmentation schemes are compared and a novel entropic similarity metric termed PSE is introduced for compound ranking. The approach is extensively tested on different compound activity classes with varying degrees of intra-class structural diversity and produces promising results in these calculations, comparable to similarity searching using state-of-the-art fingerprints. These results demonstrate the potential of randomly generated fragments for the detection of structure-activity relationships.
Furthermore, a methodology to analyze random fragment populations at the molecular level of detail is introduced. It determines conditional probability relationships between fragments. Random fragment profiles are generated for an arbitrary set of molecules, and a frequency vector is assigned for each observed fragment. An algorithm is designed to compare frequency vectors and derive dependencies of fragment occurrence. Using calculated dependency values, random fragment populations can be organized in graphs that capture their relationships and make it possible to map fragment pathways of biologically active molecules. For sets of molecules having similar activity, unique fragment signatures, so-called Activity Class Characteristic Substructures (ACCS), are identified. Random fragment profiles are found to contain compound class-specific information and activity-specific fragment hierarchies.
In virtual screening trials, short ACCS fingerprints perform well on many compound classes when compared to more complex state-of-the-art 2D fingerprints. In order to elucidate potential reasons for the high predictive utility of ACCS a thorough systematic analysis of their distribution in active and database compounds have been carried out. This reveals that the discriminatory power of ACCS results from the rare occurrence of individual and combinations of ACCS in screening databases. Furthermore, it is shown that ACCS sets isolated from random populations are typically found to form coherent molecular cores in active compounds. Characteristic core regions are already formed by small numbers of substructures and remain stable when more fragments are added. Thus, classspecific random fragment hierarchies encode meaningful structural information, providing a structural rationale for the signature character of activity-specific fragment hierarchies. It follows that compound-class-directed structural descriptors that do not depend on the application of predefined fragmentation or design schemes can be isolated from random fragment populations.},

url = {https://hdl.handle.net/20.500.11811/3701}
}

The following license files are associated with this item:

InCopyright