Even after enormous strides have been made in genomics and structure determination, over 25% of protein structures are labeled as having unknown function. Methods for computational inference or prediction of protein function can help suggest functions for such proteins, and provide independent validation of functions proposed on the basis of sequence/structure homology. Function inferences that are successfully validated in the lab can help identify potential new drug targets, predict new applications of existing drugs, and reveal alternate functions or binding sites for existing targets. They may also predict adverse drug reactions or toxicity by identifying functional similarities between unrelated proteins.

In recent work, we have developed a method for predicting the function of a protein structure using family-specific fingerprints, amino acid packing patterns occurring in most members of a protein family but rare in the background(PDB). Using a fast and robust graph representation of protein structure and a frequent subgraph mining algorithm, we were able to build a database of fingerprints from over 150 protein structure families, search within new structures for fingerprints from the families in our database, and assign function inferences with significance scores. The method was validated by predicting families for new members, and distinguishing 20 sequence/structure-similar TIM barrel families. It was used to make several predictions for the function of structural genomics proteins, some of which have been corroborated by other computational methods, and some have been validated by subsequent functional characterization.

Here we take the first steps towards the application of fingerprints for drug design, by considering some case studies of therapeutically important protein families and function inferences. This is an important step towards our goal of integrating function inference tools into the pharmaceutical pipeline, annotating new targets by functional similarity to existing families, and ultimately structure-based design of highly target-specific ligands.