
Advanced Search »

Structure Similarity Searching: Part of the Process in 2006

The new structure similarity searching feature, part of the SciFinder Substructure Search Module (SSM), is a useful way to find new leads, generate new ideas and directions, or look for competitive information. Other applications for similarity searches are when:

Substructure searches find few or no answers
Diversity is important in compounds studied with respect to a known structure
Something similar is needed, but not too similar to a compound
It is part of the standard protocol

Using the structure building interface, scientists can create a structure query as usual and tailor it by adding filters to restrict substances and/or references. From the answer set, researchers can retrieve references, supplier data, and regulatory information for substances of interest.

Exact and substructure searches are commonly used by many chemists. Similarity structure searches will not replace these searches since the two types of searches are actually complementary. Where substructure and exact searches identify substances that match a query, similarity searching is based on the closeness of structures to the one of interest. Therefore, similarity searches may retrieve substances that were retrieved by a substructure search, but the real strength of similarity searching is that substances can be retrieved that exact and substructure searches will not return, depending on the query.

Answers from a similarity structure search are sorted or clustered by similarity scores, and the user may choose the level of similarity they desire after the search completes. The closer a structure is to the substance searched, the higher the score.

Structure similarity searches can retrieve substances that are positional isomers, have different or fewer substituents, different ring systems, etc. This example shows a SciFinder similarity search.

CAS is using an industry standard metric for measuring similarity, the Tanimoto algorithm.

simsearch7

This algorithm compares descriptors in a starting compound (query) with descriptors in each database compound. The process is very fast, taking less than 5 seconds to compare the structure of interest with the entire database. The similarity score is calculated for each compound and the scores and compounds are sorted by decreasing similarity. SciFinders implementation then returns clusters of answers with scores of 60 and higher.

So, when looking for new ideas, be sure to consider using SciFinders structure similarity searching as part of your process.

Updated 4/24/2007 5:47:17 PM