Network Analysis of Protein-Protein Interaction¶
In this mini-project, Protein-Protein interaction behaviours are analysed and find the protein that change in its status has potentially the biggest effect on the rest of the network. Please, use this notebook and report file(avaiable in github) for better understaing.
#import libraries
import pandas as pd
import networkx as nx
import matplotlib.pyplot as plt
from collections import Counter
from networkx.algorithms.community import k_clique_communities
from operator import itemgetter
#load file (DM-LC)
df = pd.read_csv('DM-LC.txt', sep= '\s+', header = None)
#df = pd.read_table('DM-LC.txt', header= None)
df.head()
#Q1. Create network (g) with Networkx
g = nx.Graph()
for i in range(len(df)):
node=df.loc[i,:][0] # first column as node
next_node=df.loc[i,:][1]# first column as node
weight=df.loc[i,:][2] # third column as edge cost/weight
g.add_weighted_edges_from([(node,next_node,weight)])
#print(node,next_node,weight)
#position nodes
pos = nx.spring_layout(g)
#calculate betweeness centrality
betCent = nx.betweenness_centrality(g, normalized=True, endpoints=True)
#node color varies with Degree
node_color = [20000.0 * g.degree(v) for v in g]
# node size varies with betweeness centrality
node_size = [v * 10000 for v in betCent.values()]
#create figure
plt.figure(figsize=(20,20))
nx.draw_networkx(g, pos=pos, with_labels=False, node_color=node_color, node_size=node_size)
plt.axis('off')
#another way to draw graph
G = nx.read_weighted_edgelist('DM-LC.txt')
nx.draw(G, node_size=10)
#Q2. Compute number of nodes, number of edges and the average degree of the network
network_info = nx.info(g)
print(network_info)
#n,k = g.order(), g.size()
#av_degree = 2*float(k)/n
#print(av_degree)
#another solution
print('No. of nodes: ', nx.number_of_nodes(g))
print('No. of edeges: ', nx.number_of_edges(g))
degrees = [deg for (node, deg) in nx.degree(g)] # The degrees of each node
avg_degree = sum(degrees) / float(len(degrees))
print('Average degree: ', avg_degree )
print("Average degree: ", 2 * nx.number_of_edges(g) / nx.number_of_nodes(g))
#Q3. Compute the density of the network
density = nx.density(g)
print("Network density:", density)
#Q4. calcualte the minimum spanning tree in g and draw it.
T = nx.minimum_spanning_tree(g)
print(nx.info(T))
#list(T.edges(data=False))
pos = nx.spring_layout(T)
plt.figure(figsize=(20,20))
nx.draw_networkx(T, pos=pos, with_labels=False,
node_color='b',
node_size= 30 )
plt.axis('off')
#Q5. Draw the degree distribution histogram.
#take degrees from network(g)
degree_sequence = sorted([d for n, d in g.degree()], reverse=True)
#print(degree_sequence)
#count degree frequency
degreeCount = Counter(degree_sequence)
#print(degreeCount)
#separate degree for x-axis and its counts for y-axis
deg, cnt = zip(*degreeCount.items())
#print(deg,cnt)
#create figure
fig, ax = plt.subplots()
#ax.plot(deg, cnt, 'ro-')
ax.bar(deg, cnt, width=0.50, color='b') #creates bar chart graph of degree showing no. of nodes
plt.title("Degree Distribution in Barchart")
plt.xlabel("Degree")
plt.ylabel("Nodes Count")
plt.show()
'''Alternative way to plot histrogam of the degree in netowrk (g) by using 'hist' fucntion in matplotlib'''
#create histogram for degree
plt.hist(degree_sequence, bins='auto') #auto bin size is used
plt.title("Degree Distribution Histogram")
plt.xlabel("Degree")
plt.ylabel("Nodes Count")
plt.show()
#Compute largest connected component of the network (LC)
#lsit the components in network (g)
components = nx.connected_components(g)
#compare among components and find the one having maximun length(LC)
largest_component = max(components, key=len)
#largest_component
# Q1.draw LC
subgraph = g.subgraph(largest_component)
pos = nx.spring_layout(subgraph) # force nodes to separte
betCent = nx.betweenness_centrality(subgraph, normalized=True, endpoints=True)
node_color = [20000.0 * g.degree(v) for v in subgraph]
node_size = [v * 10000 for v in betCent.values()]
plt.figure(figsize=(20,15))
nx.draw_networkx(subgraph, pos=pos, with_labels=False,
node_color=node_color,
node_size=node_size)
plt.axis('off')
# Q2.compute diameter
diameter = nx.diameter(subgraph)
print("Diameter of LC:", diameter)
#Q3.compute center
center = nx.center(subgraph)
print('Center of LC: ', center)
# Q4.find number of clique communities with 3 nodes
cli_num = list(k_clique_communities(subgraph,3))
print('clique communities with 3 nodes: ', len(cli_num))
#Q5. node(protein) having biggest influences in LC when changing its status
#claculates eigenvector centrality in LC
eigen_cent = nx.eigenvector_centrality(subgraph)
#sorted_eigen=sorted(eigen_cent, key=eigen_centrality.get)
sorted_eigen =sorted(eigen_cent.items(), key=itemgetter(1), reverse=True)
sorted_eigen[1][0]
#calcualtes degree centrality
degree_cent = nx.algorithms.degree_centrality(subgraph)
sorted_degree = sorted(degree_cent.items(), key=itemgetter(1), reverse=True)
sorted_degree[1][0]
#calcualtes betweenness centrality in LC
betweenness_cent = nx.betweenness_centrality(subgraph, normalized=True, endpoints=True)
sorted_betweenness = sorted(betweenness_cent.items(), key= itemgetter(1), reverse=True)
sorted_betweenness[1][0]
Conclusion:¶
The diameter of LC was 18, which tells that this is the longest path of all the calculated shortest path in the network between one node to far another node. This depicts the linear size of the network (g). The centre of LC was found ['W02D3.9'] protein and the number of clique communities with 3 nodes was 31.
The protein that has the biggest effect in the network while chaining its status, the three centrality measures (degree of centrality, Eigenvector centrality and betweenness centrality ) were performed. In the LC, the protein - 'F58A4.3’ was observed for degree of centrality and eigenvector centrality while the protein - 'W02D3.9' was found as the betweenness centrality. The degree of centrality tells only about how many edges connected to a node i.e. node of the highest degree network, no matter how importance are neighbouring nodes. It doesn’t take care about it.
Unlike degree of centrality, eigenvector centrality measures the node with highest degree along the importance of neighbouring nodes. The betweenness centrality shows the important node where the majority information passes through it . Thus, 'W02D3.9' is the protein in LC that has the maximum influence in the network when changes happen in it because it is in critical node where the interaction (information) between one to another group of nodes (proteins) in LC pass through it. Thus, incase it goes inactive there is no flow of information between group of nodes in network.