1. Chapter 14 Protein Structure Classification
A classification of structures is useful for different reasons
It is helpful in understanding the evolution
Useful to describe protein fold space
Which of the possible folds exist in nature
How many different folds exists. If there exist a finite number of folds , the structure prediction problem becomes more easy
Useful to classify a new structure as being of a known/new fold
Help in understanding the relationship between structure and function
Classification makes protein 3D structure data more accessible and understandable
Classification on the basis of common fold and function informs hypothesies about how proteins evolve new functions. What is the relationship between protein fold and folding pathway?
Chapter 14 Structure classification

2. Protein Structure Classification
Mainly three systems exist for structure classification
CATH: Class - Architecture – Topology – Homologe superfamily
SCOP: Structure Classification of Proteins
Dali-FSSP and Dali-DD (Fold classification based on Structure Structure comparison of Proteins)
Most of them use Protein domains as unit for classification
Chapter 14 Structure classification

3. Chapter 14 Structure classification
Protein domains
There does not exist a general accepted definition of what a domain is, but some properties are:
A domain is part of a polypeptide chain of a protein or the whole chain
It does not need to be a contigeous region of the polypeptide chain
It can fold independently to its stabil fold
It has its own function
It contains at least one hydrophobic core
It is local compact
Chapter 14 Structure classification

4. Identifying protein domains
Different classification methods use different properties for domain definition, so the identified domains of a protein can vary with the method
The most common concepts used for domain identification are
Local compactness, a domain makes more intra-domain contacts than contacts to the residues in the remainder of the structure
It must have at least one hydrophobic core
Minimizing the number of chain-breaks needed to separate domains while also measuring the degree of contacts between the separating units
Solvent area calculation
Secondary structure elements should rarely cross between different domains
Chapter 14 Structure classification

5. An Ising model for identifying protein domains
An Ising model consists of nodes, which can be in one of several states
Each node has an initial state
The states are changed in an iteration, until all nodes belonging to a ”group” are in the same state
The changing depends on the state of its neighbour states
For domain identifiaction we have
The nodes are the residues
A group is a domain
The states are specified by numerical values
The average value of the neighbouring states (in space) are used to decide if changing, and to what
Chapter 14 Structure classification

6. An Ising model for identifying protein domains, cont’
We then must decide
The initial value
Let it be the residue number
The neighbourhood
Define a radius around the residue
How to update (change) the state of residue i
Let sit be the state of residue i after t iterations
Sit+1 = sit +k, where k is
1 if the neighbourhood has ”greater states” than residue i
-1 if the neighbourhood has ”lower states” than residue i
0 otherwise
The state of the neighbourhood depends on the states of its residues, and the distances to residue i
Must have a method for assuring termination of the iteration
Chapter 14 Structure classification

7. Chapter 14 Structure classification
Domain classes
The core of a protein is made by packing the SSEs
Two types take part in the packing , hence only three types of pairwise connections:
alpha with alpha
beta with beta
alpha with beta
All these connections may exist in a domain, but very often one of the connections dominate
The domains can therefore be classified after the dominance of a connection into different classes
Mainly alpha
Mainly beta
Alpha-beta, which can be divided into alpha/beta and alpha+beta
Chapter 14 Structure classification

8. Chapter 14 Structure classification
Folds
A fold is a special arrangement of SSEs
An open question is how many (different) folds exist in nature
Proteins in the same fold are homologous, or converged to the same fold
(Automatic) classification
SCOP is completely manually constructed
CATH partly automatically constructed
FSSP/DaliDD is fully automatically constructed
A representative set of nonredundant (unrelated) structures (less than 25% sequence identity) from PDB is constructed
Construct a distribution of the scores of all pairwise alignments between the unrelated structures
Calculate the middle value m and the standard deviation s
Two structures with scoring larger than m+2s are said to have equal folds
Chapter 14 Structure classification

9. Comparison of the different classification methods
Chapter 14 Structure classification

10. Classification by CATH
The structures are first divided into domains. Three different methods for domain identification are first used, and if not agreement, manually decision is performed.
Then the domains are classified
Class assignment: (three classes)
assign SSE to each residue (alpha, beta, loop)
represent the SSEs as sticks
count the number of residues in each SSE-type
count the numbers of contacts for alpha-alpha and beta-beta
use 2 to 4 to decide the class
Architecture assignment
Use how the SSEs are organized, independent of topology. Is performed manually
Chapter 14 Structure classification

11. Classification by CATH, cont’
Fold assignment: (Topology)
Use SSAP for comparison
Main rule: SSAP-scoring greater than 70 and 60% of the smallest structure matches the largest, is interpreted as similar fold
Homologous superfamily
Use SSAP scoring and sequence equality
Sequence families
Sequence identity greater than 35% (How to measure sequence similarity?)
Chapter 14 Structure classification

12. Classification by CATH, the procedure
Chapter 14 Structure classification