 |
 |
|
Chemical Structure Databases
Databases of chemical structures are an important tool for the pharmaceutical industry because they allow virtual screening of molecules for particular biological effects. The molecules in these databases are typically represented by atom-bond graphs. The aim of this work is to apply graph matching methods to retrieval, activity prediction and generation of new molecules. Both classic graph matching techniques and spectral methods can be applied to this problem.  Current research Clustering is an important application in large databases, and allows similar molecules to be grouped together. Previous work has looked at the clustering of graph structures based on distance measures, techniques which could be applied to chemical structure databases. Activity prediction of compounds in a database allows for virtual screening which can give an indication of which molecules may have potential for a particular application. There exist many powerful methods for achieving this on vectorial data. These methods could be extended to molecular graphs either using graph embedding methods or spectral features. The generation of new compounds which are potentially active against a particular target is an important goal in computational chemistry. The aim here is to use the information in a database of compounds and their activities to construct a generative model which can generate new compounds which may have similar activities. This is coupled to our work on generative models. Contact: Richard Wilson 
|