David

code/quantitative_index/qindex.py

@@ -0,0 +1,65 @@
+import numpy as np
+from rdkit import DataStructs
+
+metrics = {
+    'asymmetric':    DataStructs.AsymmetricSimilarity,
+    'braunblanquet': DataStructs.BulkBraunBlanquetSimilarity,
+    'cosine':        DataStructs.BulkCosineSimilarity,
+    'dice':          DataStructs.BulkDiceSimilarity,
+    'kulczynski':    DataStructs.BulkKulczynskiSimilarity,
+    'mcconnaughey':  DataStructs.BulkMcConnaugheySimilarity,
+    'rogotgoldberg': DataStructs.BulkRogotGoldbergSimilarity,
+    'russel':        DataStructs.BulkRusselSimilarity,
+    'sokal':         DataStructs.BulkSokalSimilarity,
+    'tanimoto':      DataStructs.BulkTanimotoSimilarity
+}
+
+def mean_dist_set(fps):
+    """
+    Returns an overall set dissimilarity for a list of fingerprints from RDKit, as defined in
+    "Nath A, Atkins WM. A quantitative index of substrate promiscuity. Biochemistry. 2008."
+    The fingerprints have to be the same length.
+    """
+
+    # convert to numpy array and add vectors
+    s = np.array(list(fps[0]))
+    for i in range(1,len(fps)):
+        s = s + np.array(list(fps[i]))
+
+    l = (s == len(fps)).sum() # bits common to all fps
+    k = np.count_nonzero(s) - l # on bits for at least one but not all fps
+
+    return k/(k+l)
+
+def mean_dist(fps,metric):
+    """
+    Returns the mean distances between molecules represented by a list of
+    fingerprints from RDKit, with a specified metric.
+    """
+    score = metrics[metric] # similarity score
+
+    n = len(fps)
+    md = []
+
+    for fp in fps:
+        s = sum(score(fp,fps,returnDistance=1))-1
+        md.append(s/(n-1))
+
+    return md
+
+def qindex(fps,e,metric):
+    """
+    Returns a quantitative index of substrate promiscuity (J) as defined in
+    "Nath A, Atkins WM. A quantitative index of substrate promiscuity. Biochemistry. 2008."
+    for a list of fingerprints from RDKit, with corresponding values for e = kcat/KM.
+    Metric specifies the chosen similarity score.
+    """
+    N = len(fps)
+
+    md = np.array(mean_dist(fps,metric)) # mean distances
+    nmd = md/mean_dist_set(fps) # normalized mean distances
+    e = np.array(e) # kcat/KM
+    p = e/sum(e)
+    A = -N/(sum(nmd)*np.log(N))
+
+    return A * sum(nmd * p * np.log(p))