added Conformations.dist_from_dissimilarity function

src/projet8.py

@@ -5,7 +5,11 @@ import numpy as np
 import pbxplore as pbx
 import sys
 from scipy.spatial.distance import squareform, pdist, jaccard 
-from pyclustering.cluster.kmedoids import kmedoids # for kmedoids method
+from pyclustering.cluster.kmedoids import kmedoids
+from pyclustering.cluster import cluster_visualizer
+from pyclustering.utils import read_sample
+from pyclustering.samples.definitions import FCPS_SAMPLES
+
 
 class Conformations:
     """
@@ -20,8 +24,6 @@ class Conformations:
         df : pd.DataFrame object
         Each row of the dataframe is a conformation and each column a position
         of the sequence.
-        dfnum : to compute dissimilarity with the function pdist, elements of
-        the DataFrame must be float
         """
         self.df = pd.DataFrame()
         for chain_name, chain in pbx.chains_from_files([filename]): 
@@ -51,7 +53,7 @@ class Conformations:
         If no substitution matrix is specified, use the matrix from the
         PBxplore package.
         """
-        dissimilarity = np.ndarray((self.df.shape[0],self.df.shape[0])) 
+        dissimilarity = np.zeros((self.df.shape[0],self.df.shape[0])) 
         if matrix == None:
             matrix = pd.read_table("data/PBs_substitution_matrix.dat",
                                    index_col=False, sep ='\t')
@@ -61,14 +63,30 @@ class Conformations:
         ncol = self.df.shape[1]
         nrow = self.df.shape[0]
         it1 = self.df.iterrows()
-        for i in range(1,nrow): 
+        for i in range(1,nrow): # compute each pairwise distance once only
             for j in range(i): 
                 for k in range(ncol): 
                     dissimilarity[i][j] += matrix[confs.df.loc[i,k]][confs.df.loc[j,k]]
-        dissimilarity = dissimilarity + dissimilarity.T
-
+        dissimilarity = dissimilarity + dissimilarity.T # fills the whole matrix
+        
         return dissimilarity
 
+    def dist_from_dissimilarity(diss_matrix):
+        """
+        Using the substitution matrix from the PBxplore package, the obtained
+        dissimilarity matrix has both positive and negative values. Low value
+        represent strong differences as identical PB substitution are high
+        positive value.
+        This function returns a distance matrix from the dissimilarity matrix.
+        
+        Arguments :
+            - diss_matrix : ndarray obtained from Configurations.dissimilarity
+        """
+        diss_matrix = -diss_matrix
+        diss_matrix = (diss_matrix - np.min(diss_matrix))/np.ptp(diss_matrix)
+        dist = diss_matrix * abs((np.identity(confs.df.shape[0])-1))
+        
+        return dist
 
 if __name__ == "__main__":
     if len(sys.argv) != 2: