-
Applications of Molecular Distance Measures
- Protein Classification
- Protein Alignment
- Local Matching: Geometric Hashing, Pose Clustering, and Match Augmentation
Topics in this module
In a previous module , the topic of comparing and quantifying the distance between different conformations of a given molecule was explored. Structure-based comparison is also of interest for distinct proteins, which lack the atom-by-atom correspondence necessary for RMSD calculations. In this case, an alignment is performed either based on amino acid sequence or on three-dimensional structure, and the subset of atoms successfully aligned are used as the basis for calculating conformational distance. Computing distances among entire proteins by doing a global alignment of their structures is useful for protein classification.
Protein classification
Protein classification is motivated by the notion of "descriptive biology". When faced with tremendous amounts of highly complex data, such as with the set of all proteins, one way to understand the data is by classification: the act of associating or grouping proteins into classes using certain criteria. One such criterion is protein sequence identity, where sequential similarity led to the development of phylogenetic trees and multiple sequence analyses. The same is done in protein structure classification, where the effort is to identify groups of similar proteins, with the hope that this will yield information about their biochemical function and biological purpose.
Proteins are classified by simultaneously applying a number of criteria, including sequence homology (evolutionary relatedness), function, folding motifs, structural features, and so on. The resulting hierarchies and clusters of protein structures provide a notion of the distance between two proteins and their structures. A couple of popular classification schemes are linked below. Note that a fair amount of manual annotation and classification was necessary to build these systems.
Protein alignment
The core computational problem of protein classification, using sequence or structure, is the problem of comparing two proteins. For structural classification, one method for comparison is structural alignment , which identifies an ideal superimposition between two protein structures, in order to compare them.
SSAP, Dali, Foldminer, Lock, and Geometric Hashing are algorithms which have been designed in part to align whole protein structures. Despite differences in algorithmic approach, all of these algorithms essentially evolved from the need to assign the best possible correlation between points in one structure and points in another. The problem of finding the optimal alignment is polynomial in the number of atoms in biological data, where we are assured that atoms cannot fall within a certain distance to each other (Van der Waals forces enforce this), but without this constraint the problem is exponential.
Protein alignment has been used for the classification and comparison of proteins in many existing algorithms. These include:
- Dali is a structural comparison algorithm based on pairwise distance matrices. Dali uses patterns of residue contacts, similar to contact maps described above in the intramolecular distances section, in order to align structures. The alignments are found using a randomized (Monte Carlo) search.
- FoldMiner and LOCK 2 . FoldMiner finds protein structures similar to an input structure by performing alignment the query structures secondary structure elements with proteins in its database using the LOCK 2 algorithm. LOCK 2 uses a combination of geometric hashing and dynamic programming to optimize the alignments of secondary structure elements of different proteins. Once a set of alignments to similar structures are found, motifs consisting of similar secondary structure arrangements are constructed and used to refine the similarity search.
- Sequential Structure Alignment Program (SSAP) Given two protein structures, SSAP returns a structural alignment.