Révision 4e82c425 modules/clustering.py

b/modules/clustering.py
1 1 
"""Functions to cluster structures.
2 2 

  		





  3
  3
  
    
functions:


  		





  
  4
  
    
get_rmsd: Computes the rmsd matrix of the conformers in a list of rdkit mol


  		





  
  5
  
    
    objects.


  		





  4
  6
  
    
get_labels_affty: Clusters data in affinity matrix form by assigning labels to


  		





  5
  
  
    
data points.


  		





  
  7
  
    
    data points.


  		





  6
  8
  
    
get_labels_vector: Clusters data in vectorial form by assigning labels to


  		





  7
  
  
    
data points.


  		





  
  9
  
    
    data points.


  		





  8
  10
  
    
get_clusters: Groups data-points belonging to the same cluster into arrays of


  		





  9
  
  
    
indices.


  		





  
  11
  
    
    indices.


  		





  10
  12
  
    
get_exemplars_affty: Computes the exemplars for every cluster and returns a list


  		





  11
  
  
    
of indices.


  		





  
  13
  
    
    of indices.


  		





  12
  14
  
    
plot_clusters: Plots the clustered data casting a color to every cluster.


  		





  13
  15
  
    
clustering: Directs the clustering process by calling the relevant functions.


  		





  14
  16
  
    
"""


  		





  ......




  21
  23
  
    

  		





  22
  24
  
    

  		





  23
  25
  
    
def get_rmsd(mol_list: list, remove_Hs="c"):


  		





  24
  
  
    
    """Computes the rmsd matrix of the conformers in a rdkit mol object.


  		





  
  26
  
    
    """Computes the rmsd matrix of the conformers in a list of rdkit mol objects


  		





  25
  27
  
    

  		





  26
  28
  
    
    @param mol_list: list of rdkit mol objects containing the conformers.


  		





  27
  29
  
    
    @param remove_Hs: bool or str,


  		





  ......




  92
  94
  
    
    @return: list of cluster labels. Every data point is assigned a number


  		





  93
  95
  
    
    corresponding to the cluster it belongs to.


  		





  94
  96
  
    
    """


  		





  
  97
  
    
    # TODO Implement it.


  		





  95
  98
  
    
    return []


  		





  96
  99
  
    

  		





  97
  100
  
    

  		





  ......




  115
  118
  
    
    distances between points, RMSD Matrix, etc.) shape: [n_points, n_points].


  		





  116
  119
  
    
    @param clusters: tuple of arrays. Every array contains the indices (relative


  		





  117
  120
  
    
    to the affinity matrix) of the data points belonging to the same cluster.


  		





  118
  
  
    
    @return: list of indices (relative to the affinity matrix) exemplars for


  		





  119
  
  
    
    every cluster.


  		





  
  121
  
    
    @return: list of indices (relative to the affinity matrix) of the exemplars


  		





  
  122
  
    
    for every cluster.


  		





  
  123
  
    

  		





  
  124
  
    
    This function finds the exemplars of already clusterized data. It does


  		





  
  125
  
    
    that by (i) building a rmsd matrix for each existing cluster with the values


  		





  
  126
  
    
    of the total RMSD matrix (ii) carrying out an actual clustering for each


  		





  
  127
  
    
    cluster-specific matrix using a set of parameters (large negative value of


  		





  
  128
  
    
    preference) such that it always finds only one cluster and (iii) it then


  		





  
  129
  
    
    calculates the exemplar for the matrix.


  		





  120
  130
  
    
    """


  		





  121
  131
  
    
    from sklearn.cluster import AffinityPropagation


  		





  
  132
  
    
    # Splits Total RMSD matrix into cluster-specific RMSD matrices.


  		





  122
  133
  
    
    clust_affty_mtcs = tuple(affty_mtx[np.ix_(clust, clust)]


  		





  123
  134
  
    
                             for clust in clusters)


  		





  124
  135
  
    
    exemplars = []


  		





  
  136
  
    
    # Carries out the forced-to-converge-to-1 clustering for each already


  		





  
  137
  
    
    # existing cluster rmsd matrix and calculates the exemplar.


  		





  125
  138
  
    
    for i, mtx in enumerate(clust_affty_mtcs):


  		





  126
  139
  
    
        pref = -1e6 * np.max(np.abs(mtx))


  		





  127
  140
  
    
        af = AffinityPropagation(affinity='precomputed', preference=pref,


  		












Formats disponibles :
  Unified diff