CDK2 Clustering
===============

1. The simplest way to cluster structures based upon residue properties is to use the :doc:`polyphony_Ete2_tree` script. This will generate an interactive tree in Ete and write out a Newick format file. This is a standard tree text format and can be read in using a number of programs. Here is a dendrogram produced using `FigTree <http://tree.bio.ed.ac.uk/software/figtree/>`_. It shows the CDK2 chains clustered by Tanimoto similarity of the residue interaction fingerprints generated from protein-protein interaction database `Piccolo <http://www-cryst.bioc.cam.ac.uk/~richard/piccolo/piccolo.php>`_ (-p ppi option). 


.. image:: images/cdk2/ppi-tanimoto.png
    :scale: 50 %
    :align: center
    
It shows the chains with no biologically relivant protein contacts in blue and those interacting, usually with cyclin a, in yellow.
    
2. Groups can be generated from the hierarchical clustering. This isn't done in a very sophisticated way. One way is to use: 

.. automethod:: Polyphony.Trees.Tree.biggest_left_right_others

This method works well to split CDK2 chains into monomers and multimers (blue and yellow in above diagram).

e.g.

.. code-block:: python

    from Polyphony.Trees import Tree
    from Polyphony.Comparison_Matrices import Structure_Matrix
    from Polyphony.Structural_Alignment import Structural_Alignment
    from Polyphony.Utils import Properties
    from Polyphony.Plotting import plot_alignment_properties
        
    ## Main program
    
    # Alignment file etc.
    filename = "clust_1HCK_A_95.fasta"
    update = False
    property = "backbone"
    
    # Create structural alignment
    aligned = Structural_Alignment()
    aligned.add_alignment(filename)
    
    # Get/calculate selected property
    properties = Properties()
    array = properties.get_array(property, aligned, update)
    
    # Cluster by protein-protein interactions
    clust_array = properties.get_array("ppi", aligned, update)
    structmat = Structure_Matrix(clust_array)
    tree = Tree(structmat.data, structmat.get_labels())
    
    # Group into active and inactive
    groups = tree.biggest_left_right_others(1)
    unbound = groups[0]
    bound = groups[1]
    
    # Plot curvature and torsion, coloured by group
    xlabels = aligned.get_consensus_sequence()
    plot_alignment_properties(array.data, property, xlabels, array.dim_names, colour_groups=[unbound, bound])
    

.. image:: images/cdk2/dfg-curvature-torsion.png
    :scale: 25 %
    :align: center
    
The same region can be visualised in 3D using in PyMol. Type pymol -R in one console window and run the following in another, preferably in ipython.

.. code-block:: python

    from Polyphony.Pymol import Pymol_Viz
    
    # Get started and load representative structure into PyMol
    cdk2 = Pymol_Viz("clust_1HCK_A_95.fasta","cdk2")
    
    # Group in to bound and unbound structures
    groups = cdk2.group_biggest_clusters(property="ppi")
    unbound = groups[0]
    bound = groups[1]
    
    # Convert the lists of alignment index numbers to chain ids
    unbound_ids = cdk2.ids_for_groups(unbound)
    bound_ids = cdk2.ids_for_groups(bound)
    
    # Find preferred segment of structurally conserved residues. Try experimenting with the cutoff. The default value of 0.5 can be a bit high.
    cdk2.colour_conserved_segments(0.1)
    
    # Load all structures in each group into a PyMol group allowing them to be coloured separately
    # They then aligned using your chosen segment. This will take a few minutes
    cdk2.load_structures(id_list=unbound_ids, segment=[101,111], pymol_group="unbound")
    cdk2.load_structures(id_list=bound_ids, segment=[101,111], pymol_group="bound")


You can then colour each group separately, view as ribbons etc.

.. image:: images/cdk2/dfg-bound-unbound.png
    :scale: 25 %
    :align: center