Analysis of Random Fragment Profiles

Fragment-type descriptors are effective and widely
used tools in
computer-aided drug discovery. This book introduces a
novel method that departs from traditional fragment
design schemes. Fragment profiles of molecules are
generated by random deletion of bonds. Such profiles
can be mined for substructures associated with
different compound classes. For the analysis of
molecular similarity relationships, the profiles are
quantitatively compared using entropy-based metrics.
Similarity searching experiments on compound classes
with varying structural diversity produce promising
results, in respect to traditional fingerprints. In a
further step, random profiles were minded for
dependency relationships of fragment co-occurrence,
which enables to isolate substructure pathways of
biological active compounds. So-called Activity Class
Characteristic Substructures (ACCS) can be identified
for molecules with similar activity. In virtual
screening, short ACCS fingerprints perform comparable
to state-of-the-art fingerprints of higher
complexity. Therefore, the analysis of randomly
generated fragment profiles open up new
possibilities for the design of fragment-type
descriptors.

Autorentext

Jose Batista, Dr. rer. nat.: Degree in Computer Science,specialised in Chemoinformatics at the B-IT, Dept. of LifeScience Informatics, University of Bonn. Research scientist atJADO Technologies, Dresden

Klappentext

Fragment-type descriptors are effective and widely used tools in computer-aided drug discovery. This book introduces a novel method that departs from traditional fragment design schemes. Fragment profiles of molecules are generated by random deletion of bonds. Such profiles can be mined for substructures associated with different compound classes. For the analysis of molecular similarity relationships, the profiles are quantitatively compared using entropy-based metrics. Similarity searching experiments on compound classes with varying structural diversity produce promising results, in respect to traditional fingerprints. In a further step, random profiles were minded for dependency relationships of fragment co-occurrence, which enables to isolate substructure pathways of biological active compounds. So-called Activity Class Characteristic Substructures (ACCS) can be identified for molecules with similar activity. In virtual screening, short ACCS fingerprints perform comparable to state-of-the-art fingerprints of higher complexity. Therefore, the analysis of randomly generated fragment profiles open up new possibilities for the design of fragment-type descriptors.