A common way to gain insight into the underlying genetic components of a disease is through running an experiment with a DNA Microarray. A microarray can detect the levels of thousands of genes simultaneously. More details on the way these datasets are created from biological experiments can be found here.
First, you will be analyzing the results of a microarray experiment related to breast and ovarian cancer, found in the publication Gene Expression Patterns in Ovarian Carcinomas, by Schaener et al. Mol. Biol. Cell, Nov 1 2003, vol 14, no 11. The following file contains a subset of their experimental data.
Create a Jupyter notebook, and read this file into a Pandas DataFrame. The gene expression data is separated by tabs. The first row specifies the labels of each column. In this file, each row shows the expression levels for one of 978 different genes. The first column lists the name of the gene, and the remaining columns show results for 16 samples, one from each patient with cancer.
An individual value in this dataset is a number representing the normalized logarithm of how a particular gene was expressed in a particular sample in comparison to a normal sample.
To visually see and compare trends across the whole microarray experiment, a heatmap plot is commonly used.
First, melt the data so that you have three columns, one for the NAME, one for the Sample, and one for the value.
Reset the Sample column to be Categorical based on the order in the original data frame, since this is different than alphabetical.
Construct a heatmap where red indicates that this gene is expressed much less than expected in control conditions (negative numbers), and green indicates that this gene is expressed much more than expected (positive numbers).
Discuss any patterns you notice in the figure.
Now we will need to cluster the genes in this data.
Use K-means clustering to separate the genes into 2 clusters.
Reorganize your data by making the NAME column Categorical based on these clusters, such that the genes in one cluster are placed above the genes in the other cluster.
Draw a plot with this new data.
Report on any patterns you notice in the figure, and compare the figure you created with those found in the original scientific paper.
Next, we look at a different dataset, from A Global Map of Human Gene Expression by Lukk et al. Nature Biotechnology April 2010. This comprehensive survey tried to understand the structure of genetic expression through examining 5,372 human samples representing 369 different cell and tissue types, disease states and cell lines. Each cell is associated with 22,283 microarray experiments. We will use their data and replicate their study to visualize the cell types using PCA.
The original data is transposed from how we need it to use our sklearn PCA library, so use the following lines of code to write a new reformated file.
df = pd.read_csv("data/hgu133a_rma_okFiles_080619_MAGETAB.csv", sep='\t', header=None).T df.columns = df.iloc[0] df.drop(0,inplace=True) df.to_csv("data/hgu133a_rma_okFiles_080619_MAGETAB_fixed.csv", index=False)
Open a new notebook, and load the fixed data as a Pandas DataFrame. The first two columns are information about the sample, and the remaining columns are the microarray experiments.
Use PCA to find two principal components.
Draw a plot of the data on these two components.
Describe how much variance is explained through using these components.
Finally, load up the data file E-MTAB-62.sdrf.txt that includes the cell types for each of the samples used in the study.
E-MTAB-62.sdrf.txt
Color your plot above using the Characteristics[Blood/NonBlood meta-groups] column.
Characteristics[Blood/NonBlood meta-groups]
Then make another plot using the Factor Value[4 meta-groups] column for the color.
Factor Value[4 meta-groups]
Report on any patterns you notice in the figures.
Compare the figures you created with those found in the original scientific paper.