Abstract
It has been established that protein function is largely determined by its structure, which in turn is determined by its sequence. Proteins with similar structure or function often share sequence motifs. Such conserved sequence motifs have been used to predict protein functions. Previous work in our group has shown that classifiers for assigning proteins to one of several functional families using a combination of motifs often outperform those that use a single characteristic motif for each family. Since protein function is largely determined by structure, it is possible to classify proteins according to their function.

In this talk, I will describe an approach of data-driven construction of protein function classifiers using sequence and/or structural features of proteins. I will compare the performance of classifiers that use sequence features alone, with those that use structural features alone, and with those that useboth sequence and structrural features. Preliminary results show that classifiers that use both sequence and structural features outperform those that use sequence or structural features alone.