Scribd
Upload
3 views2 pages

Graph-Based Protein Clustering Analysis

The document loads taxonomic data, creates an adjacency matrix showing co-occurrences of taxa, constructs a graph from the matrix, embeds the graph nodes in 2D using t-SNE, clusters the embedded nodes using DBSCAN, and plots the results.

Uploaded by

Hugo
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd
3 views2 pages

Graph-Based Protein Clustering Analysis

The document loads taxonomic data, creates an adjacency matrix showing co-occurrences of taxa, constructs a graph from the matrix, embeds the graph nodes in 2D using t-SNE, clusters the embedded nodes using DBSCAN, and plots the results.

Uploaded by

Hugo
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd

### Loading the file ###

import pandas as pd
f="C:/MP-CHEW/CHEW/cycle_2/[Link]"
df=pd.read_csv(f,sep="\t")#.sample(2000) #you can just sample here for testing
purposes
df["proteins"]=df["proteins"].[Link](", ")
edf=[Link]("proteins")
edf["OX"]=edf["proteins"].[Link]("_").apply(lambda x: "_".join(x[-3:])) #these
are the taxonomies (GTDB taxid instead of NCBI)

#Data reduction: trim taxa based on frequency


s=[Link]("OX").size()
edf=edf[edf["OX"].isin(s[s>5].index)]
ut=edf["OX"].drop_duplicates()

### create adjacency matrix ###


dfm =edf[["u_ix","OX"]].merge(edf[["u_ix","OX"]],on="u_ix").query("OX_x != OX_y")
out=[Link](dfm["OX_x"],dfm["OX_y"])

#Data reduction: trim crosstab based on minimum adjacency?


# q=out[[Link]()>2].index
# out=[Link][q,q]

### Graph construction

import networkx as nx
from node2vec import Node2Vec

graph=nx.from_pandas_adjacency(out)
node2vec=Node2Vec(graph,dimensions=10,walk_length=5,num_walks=20,workers=4) #not
sure what parameters I should select here
model=[Link](window=10,min_count=1)

### 2D embedding
from [Link] import TSNE
from [Link] import PCA
import numpy as np
import [Link] as plt
# Retrieve node embeddings and corresponding subjects
node_targets = [Link].index_to_key # list of node IDs
node_embeddings = (
[Link]
) # [Link] of size number of nodes times embeddings dimensionality

trans = TSNE(n_components=2) # or PCA


node_embeddings_2d = trans.fit_transform(node_embeddings)

### Clustering

from [Link] import DBSCAN


sw=[Link][node_targets]
clustering = DBSCAN(eps=2,
min_samples=1).fit(node_embeddings_2d)#,sample_weight=sw) #distance would need
optimization?
clusters=clustering.labels_
uc=[Link](clusters)
cluster_count=len(uc)

#plot clusters
import seaborn as sns
node_colors=[Link](sns.color_palette("Spectral",n_colors=cluster_count))
[clusters]
[Link]( node_embeddings_2d[:, 0],
node_embeddings_2d[:, 1],
c=node_colors,)

#plot intensity
[Link]( node_embeddings_2d[:, 0],
node_embeddings_2d[:, 1],
c=[Link]([Link][node_targets].values)
,cmap='Spectral',s=0.2)

You might also like