What is a reasonable chemical structure representation?  A typical chemist reply is: one that I can accurately and unambiguously interpret.  A practical answer, however, a chemist interpretation and a computer interpretation may be at distinct odds when defining what �reasonable� means.  There are many implicit vs. explicit aspects to a chemist interpretation lost when interpreting a chemical structure.  There are missing formal charges (e.g., nitro groups represented by �*N(=O)O�), implied resonance by using double bonds (e.g., carboxylic acid represented by �*C(=O)=O� rather than �*C(=O)O�), and so on.  A chemist might overlook and implicitly understand and correct these aspects in their understanding, but a computer is often confused and happy to reject a chemical structure as being unreasonable without substantial chemical intuition programmed into the interpretation.  Alternatively, a chemical structure may seem completely reasonable by all known valence rules, etc., to a compute but is completely reject by a chemist as being impossible to exist (e.g., �OOOOOOOOO�, poly-peroxides).  This presentation will outline new strategies being explored at PubChem to determine whether a chemical structure is unreasonable or impractical.  The basic approach is to take advantage of a comprehensive survey of first, second, and third order nearest-neighbor environments and develop a self-learning and automated statistical-based approach to reliably predict the likelihood a chemical is reasonable (or its better representation!) without the need for developing and maintaining a set of transforms (e.g., SMIRKS).

New strategies to normalize chemical structure representations and weed-out impractical small molecules

243rd National Meeting (2012)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.