Development and Improvement of Tools and Algorithms for the Problem of Atom Type Perception and for the Assessment of Protein-Ligand-Complex Geometries
In context of the present work, a scoring function for protein-ligand complexes has been developed, not aimed at affinity prediction, but rather a good recognition rate of near native geometries. The developed program DSX makes use of the same formalism as the knowledge-based scoring function DrugSc...
|PDF Full Text
No Tags, Be the first to tag this record!
|In context of the present work, a scoring function for protein-ligand complexes has been developed, not aimed at affinity prediction, but rather a good recognition rate of near native geometries. The developed program DSX makes use of the same formalism as the knowledge-based scoring function DrugScore, hence using the knowledge from crystallographic databases and atom-type specific distance-dependent distribution functions. It is based on newly defined atom-types. Additionally, the program is augmented by two novel potentials which evaluate the torsion angles and (de-)solvation effects. Validation of DSX is based on a literature-known, comprehensive data-set that allows for comparison with other popular scoring functions.
DSX is intended for the recognition of near-native binding modes. In this important task, DSX outperforms the competitors, but is also among the best scoring functions regarding the ranking of different compounds.
Another essential step in the development of DSX was the automatical assignment of the new atom types. A powerful programming framework was implemented to fulfill this task. Validation was done on a literature-known data-set and showed superior efficiency and quality compared to similar programs where this data was available. The front-end fconv was developed to share this functionality with the scientific community. Multiple features useful in computational drug-design workflows are also included and fconv was made freely available as Open Source Project.
Based on the developed potentials for DSX, a number of further applications was created and impemented:
The program HotspotsX calculates favorable interaction fields in protein binding pockets that can be used as a starting point for pharmacophoric models and that indicate possible directions for the optimization of lead structures.
The program DSFP calculates scores based on fingerprints for given binding geometries. These fingerprints are compared with reference fingerprints that are derived from DSX interactions in known crystal structures of the particular target.
Finally, the program DSX_wat was developed to predict stable water networks within a binding pocket. DSX interaction fields are used to calculate the putative water positions.