Visualizing RNAseq Data

Getting Started

import pandas as pd
import numpy as np
import seaborn as sns
import glob
import matplotlib.pyplot as plt
sns.set_context('paper')
sns.set_style("whitegrid")

Volcano Plot

This is what we aim to reproduce basing on the file volcano_data.tsv. Let's read the volcano_data.tsv file into a pandas dataframe. glob.glob('C:\\Users\duan\Desktop\PythonDataProcessingVisualization\*.tsv') # get a list of files in your directory ending in .tsv

vol = pd.read_csv('C:\\Users\\duan\\Desktop\\PythonDataProcessingVisualization\\KOvsWTdiffExp.tsv', sep='\t')

inspect the 'head' of the file

vol.head()

vol.shape

(4900, 6)

Notice that the qvalues in the file need to be log-transformed to match the figure Create a new column in the vol dataframe where 'log10_q' = -log_base_10(qval) for each gene...the numpy function np.log10 is helpful

vol['log10_q'] = -np.log10(vol['padj']) #calculate the -log2 of qvalues

output a summary of the data (use the .describe() function)

vol.describe()

We want to plot and color the genes that increase in knockout, decrease with knockout, and show no significant change (q-val > 0.05). Let's categorize our data

create a new column called 'data_category' with entires 'increases in Knockout', 'decreases in Knockout', and 'not significant' set these values appropriately for each gene hint - inspect the 'log2FoldChange' value or the 'padj' fields to determine each case

sns.scatterplot(data=vol, y='log10_q', x='log2FoldChange', hue='data_category', legend='brief',
                palette={'not significant':'grey', 'increases in Knockout':"red", 'decreases in Knockout':"blue"}, 
                edgecolor='grey',s=80, linewidth=0.25)

save a list of the significant genes and their qvalues (any gene_ids with qval<0.05), and output this list to a .csv file

sig_genes = vol.loc[vol['padj']<.05, ['geneID', 'padj', 'log2FoldChange']]
sig_genes.shape

(2582, 3)

sig_genes.head(8)

Save significant genes to a csv file

sig_genes.to_csv('C:\\Users\duan\Desktop\PythonDataProcessingVisualization\significant_hits.csv')

Heatmap

Next, we are going to reproduce a heatmap below. This clustered heat map aims to show groups of genes whose transcript levels are coordinated across age or mutant background

The plotting will be based on rpkm.tsv which contains ~600 significant genes we want to inspect.

read in the 'rpkm.tsv' file as a pandas dataframe, save it as dataset

dataset = pd.read_csv('C:\\Users\duan\Desktop\PythonDataProcessingVisualization\\rpkm.tsv', sep='\t')

briefly inspect the dataframe for shape, general entries, and summary statistics

dataset.shape

(600,7)

dataset.head()

dataset.describe()

People often plot using 'row median centered' each gene This means they divided each row by the median value across that entire row. They also log-transformed that result

Calculate the median value for each gene (row)

row_medians = dataset.median(axis=1,numeric_only=True)

Now create a copy of the dataset, and save it as dataset_row_norm Row median center and log2-transform each gene in dataset_row_norm

dataset_row_norm = dataset.copy() #make a copy of the dataset so we can manipulate it
for col in dataset.columns[1:]:
    dataset_row_norm[col] = np.log2((dataset[col]+0.1)/(row_medians+0.1))

look at the summary statistics now Also, look at a few random gene rows and make sure the result in sensible Finally, look at the .head() to see how it is indexed

dataset_row_norm.describe()

dataset_row_norm.iloc[200,1:].median() #test some random rows, make sure the median value is 0

-0.04326176484815142

dataset_row_norm.head()

change the indexing to use the gene name instead of the row number

dataset_row_norm = dataset_row_norm.set_index('geneID')

Now let's plot the full dataset. Use seaborn clustermap to generate a heat map and cluster each row. Look at the seaborn clustermap documentation to figure out what arguments to pass

import fastcluster

sns.clustermap(dataset_row_norm, row_cluster=True, col_cluster=False, cmap="RdBu_r", figsize=(10,10))

Another Example

sometimes you need additional work to make a nice heatmap. See the example below:

Read in the example data

dataset = pd.read_csv('C:\\Users\duan\Desktop\PythonDataProcessingVisualization\\TPM_reads_raw.tsv', sep='\t')

Examine the data

dataset.shape

(5823, 36)

dataset.head()

dataset.describe()

Prepare the data for heatmap plotting

row_medians = dataset.median(axis=1,numeric_only=True) #calculate the median of each row
dataset_row_norm = dataset.copy() #make a copy of the dataset so we can manipulate it
for col in dataset.columns[1:]:
    dataset_row_norm[col] = np.log2((dataset[col]+0.1)/(row_medians+0.1))
dataset_row_norm = dataset_row_norm.set_index('gene')

Heatmap plotting

sns.clustermap(dataset_row_norm, row_cluster=True, col_cluster=False, cmap="RdBu_r", figsize=(10,10))

It is necessary to 'zoom' in on the genes that showed significant changes. Sometimes people 'capped' their fold changes at -1.5 and +1.5, we'll do the same.

Start by copying our row-normalized dataframe to a new dataframe and setting all values greater than 1.5 to 1.5, and all less than -1.5 to -1.5 Save this as capped_row_norm_dataset

capped_row_norm_dataset = dataset_row_norm.copy()
capped_row_norm_dataset[capped_row_norm_dataset>1.5] = 1.5
capped_row_norm_dataset[capped_row_norm_dataset<-1.5] = -1.5

Look at the summary statistics to see if this worked

Look at the summary statistics to see if this worked

Read volcano plot file

vol = pd.read_csv('C:\\Users\\duan\\Desktop\\PythonDataProcessingVisualization\\volcano_data.tsv', sep='\t')

Inspect volcano plot file

vol.head()

vol.shape

(13547, 4)

Prepare for volcano plotting

vol['log10_q'] = -np.log10(vol['qval']) #calculate the -log2 of qvalues

vol.describe()

vol.loc[vol['log2foldchange']>0,'data_category'] = 'increases with age'
vol.loc[vol['log2foldchange']<0,'data_category'] = 'decreases with age'
vol.loc[vol['qval']>0.05,'data_category'] = 'not significant'

Volcano plotting

sns.scatterplot(data=vol, y='log10_q', x='log2foldchange', hue='data_category', legend='brief',
                palette={'not significant':'grey', 'increases with age':"red", 'decreases with age':"blue"}, 
                edgecolor='grey',s=80, linewidth=0.25)

Identify significant genes

sig_genes = vol.loc[vol['qval']<.05, ['gene_id', 'qval', 'log2foldchange']]

Now we need to just pull out the genes that show significant age-dependence Make a list of all the gene names that are in both dataset_row_norm and the list of significant genes (sig_genes) from above. Save this list as genes_to_cluster

#get the genenames that are the same between the datasets
genes_to_cluster = [gene for gene in capped_row_norm_dataset.index if gene in sig_genes.values] 

use the .loc function to pull out just the rows of genes we want to cluster, and see how many genes that is

capped_row_norm_dataset.loc[genes_to_cluster, :].shape

(1731, 35)

Almost there - now make your heatmap!

sns.clustermap(capped_row_norm_dataset.loc[genes_to_cluster, :], row_cluster=True, col_cluster=False, cmap="RdBu_r", figsize=(8,8))