Day 3 Exercise

In this exercise, we use a common gene expression dataset for acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML) to practically implement the concepts taught during this part of the lecture. This data has gene expression values from 72 samples (ALL = 47 and AML = 25) for 7129 genes.

The expression matrix is ALLcancerdata.txt.

There is also a file containing sample annotations: ALLannotation.txt

• Samples: These are the patient sample numbers/IDs.
• ALL.AML: Type of tumor, whether the patient had acute myeloid leukemia (AML) or acute lymphoblastic leukemia (ALL).
• BM.PB: Whether the sample was taken from bone marrow (BM) or from peripheral blood (PB).
• T.B.cell: ALL arises from two different types of lymphocytes (T-cell and B-cell). This specifies which for the ALL patients; it is NA for the AML samples.

What you need to do for this exercise is:

1. read the expression matrix and the annotation table. For the annotation, we only use “ALL.AML” annotation while remove all the other threes (or you can keep all annotations if you want to try). Try to process and clean the expression matrix and the annotation data frame as you did on Tuesday.

2. There are still quite a lot of negative values in it (bad data points from the microarray). You need to process the matrix to remove genes with too many negative values and also too many small values.
   • fix the lowest and highest values to 100 and 16000 respectively
   • remove genes if max/min <= 5 or max - min <= 500
   • perform a log10() transformation on the original matrix. Before doing this, you can first check the distribution of the matrix before and after log10 transformation.

With the log10-transformed matrix, you can perform following analysis:

• PCA analysis and use different colors for AML and ALL samples
• perfrom k-means clustering with 2,3,4 groups and compare to AML/ALL annotation
• make heatmap with the complete matrix with AML/ALL as top annotation (use ComplexHeatmap package)
• make heatmap with top 1000 genes with highest variance/IQR/cv (which heatmap is better in sense of separating AML/ALL samples?)