Comparative study of probability models for compound similarity searching

Abstract

The quality of a chemical retrieval system heavily depends on its molecular similarity function which returns a similarity measurement between the target compound and each molecule in the collection. Compounds are sorted according to their similarity values with the query and those with high ranks are returned to the users. Most current chemical retrieval systems use the vector space model for similarity calculation. In this paper, the use of probability of relevance for compound retrieval is explored. It reports on the effectiveness of the probability model for compound similarity searching by using Binary Independence Model and Binary Dependence Madel on two different databases. The result based on fusion of queries for both models is also discussed. The results show that in all cases, Binary Independence Retrieval model performed better than Binary Dependence model. It is also found that fusion does not give better results than the un-fused queries.