Computational Chemistry Robots

Abstract

Millions of compounds are now Openly available (e.g. PubChem) and we describe the automatic computation of their geometries and properties. Using completely automatic procedures, based on modular components and workflow technology (Taverna) we can:

* extract structures from 3D databases or crystallographic publications * determine a cost-effective level of theory * optimise ground state geometry and calculate properties * disseminate the results Openly.

Although error rates are low their management must be completely robotic.

By using spare capacity (Condor) we have calculated 250,000 molecules at PM5 (MOPAC) and over 10000 at B3LYP/63-1G* (GAMESS), and analysed the data robotically, including: * variability between crystallographic experiment and levels of theory * geometric variability within instances of a given functional group * detection of molecular features that give rise to serious errors or pathological computation.

The results in our WorldWideMolecularMatrix (WWMM, http://wwmm.ch.cam.ac.uk) are Openly available in our DSpace repository (http://www.dspace.cam.ac.uk/handle/1810/724).