0.5 Protein classification, local alignment, and motifs  (Page 5/5)

Designing effective motifs

Given a motif representing a specific biochemical function, the task of any algorithm seeking to predict protein functions is to identify other proteins with similar biological functions. This is immediately affected by how well the motif represents the biological function - protein structures are flexible, dynamic objects, and any computational representation of these objects is likely to be inaccurate in some manner. In particular, since it is the computational representation that is compared, the fidelity of representation is essential for the effectiveness of the structural comparison approach.

An effective motif must simultaneously fulfill two criteria:

• The motif must be sensitive
• The motif must be specific

Motifs must maintain geometric and chemical similarity, in respect to the characteristics compared by the comparison algorithm, to functional homologs. This ensures that algorithms searching for the motif identify proteins with similar function.

Sensitivity is measured as the proportion of acceptable matches to functional homologs relative to the total number of functional homologs. Specific motifs must maintain geometric and chemical dissimilarity, in respect to the characteristics compared by the comparison algorithm, to functionally unrelated proteins. This ensures that algorithms searching for the motif accidentally identify less matches to functionally unrelated proteins. Specificity is measured as the proportion of rejected matches to functionally unrelated proteins, relative to all functionally unrelated proteins. The ideally effective motif is 100% sensitive and 100% specific.

Under most conditions, effective motifs have been designed by hand. However, a few recently studies exist which examine the relationship between elements of the motif and the sensitivity and specificity of the motif. These studies resulted in the design of MultiBind , an algorithm which identifies conserved binding patterns among functionally homologous active sites, in order to generate motifs which represent only conserved binding patterns. Another study of motif design produced Geometric Sieving , an algorithm for maximizing the Geometric Uniqueness of motifs from functionally unrelated proteins.

Putting together effective motif design, an efficient matching algorithm, and a statistical scoring of the results can lead to an automated functional annotation pipeline for proteins.

Recommended reading:

• Chen, B.Y. (2005). Algorithms for Structural Comparison and Statistical Analysis of 3D Protein Motifs . Proceedings of the Pacific Symposium on Biocomputing 2005, pp. 334-345.
• Chen, B.Y. (2006). Geometric Sieving: Automated Distributed Optimization of 3D Motifs for Protein Function Prediction . Proceedings of the Tenth Annual Conference on Research in Computational Molecular Biology (RECOMB 2006), pp. 500-512.
• Yao, H., Kristensen, D., Mihalek, I., Sowa, M., Shaw, Ch., Kimmel, M., Kavraki, L. E., and Lichtarge, O. (2003). An Accurate, Sensitive, and Scalable Method to Identify Functional Sites in Protein Structures . Journal of Molecular Biology, 325:255-261.
• David M. Kristensen, Brian Y. Chen, Viacheslav Y. Fofanov, R. Matthew Ward, A. Martin Lisewski, Marek Kimmel, Lydia E. Kavraki, and Olivier Lichtarge. (2006). Recurrent Use of Evolutionary Importance for Functional Annotation of Proteins Based on Local Structural Similarity . Protein Science, in press.