FLASHFLOOD: A 3D field-based similarity search and alignment method for flexible molecules

A three-dimensional field-based similarity search and alignment method for flexible molecules is introduced. The conformational space of a flexible mo lecule is represented in terms of fragments and torsional angles of allowed conformations. A user-definable property field is used to compute features of fragment pairs. Features are generalizations of CoMMA descriptors (Silv erman, B.D. and Platt, D.E., J. Med. Chem., 39 (1996) 2129.) that character ize local regions of the property field by its local moments. The features are invariant under coordinate system transformations. Features taken from a query molecule are used to form alignments with fragment pairs in the dat abase. An assembly algorithm is then used to merge the fragment pairs into full structures, aligned to the query. Key to the method is the use of a co ntext adaptive descriptor scaling procedure as the basis for similarity. Th is allows the user to tune the weights of the various feature components ba sed on examples relevant to the particular context under investigation. The property fields may range from simple, phenomenological fields, to fields derived from quantum mechanical calculations. We apply the method to the di hydrofolate/methotrexate benchmark system, and show that when one injects r elevant contextual information into the descriptor scaling procedure, bette r results are obtained more efficiently. We also show how the method works and include computer times for a query from a database that represents appr oximately 23 million conformers of seventeen flexible molecules.