Transcript ALS - BioQUEST

BioQUEST / SCALE-IT Module
From Omics Data
to Knowledge
Case 1: Microarrays
Namyong Lee
Matthew Macauley
Sumona Mondal
Fusheng Tang
Minnesota State University, Mankato
Clemson University
Clarkson University
University of Arkansas, Little Rock
Goals


Provide a guideline for teachers in different
disciplines to explore different -omics data.
The instructor will guide the students
through a tutorial of the experimental
process, including: data retrieval, statistical
design and analysis, biological analysis,
and model validation.
Module Outline
1.
2.
3.
4.
5.
6.
Introduce Microarray and RNAseq technology.
Locate available public expression data
Formulate questions from the dataset.
Design computational and statistical
experiments.
Interpret biological significance of identified
genes. (UniProt, IntAct, and Reactome will be
used.)
Validate the biological model (using ATLAS).
Step 1: Introduce gene expression
and microarray and RNAseq
technology.
 What is gene expression?

How is gene expression measured?
Introduce microarrays and RNAseq. Compare
and contrast these two.
Step 2: Locate available public
expression data

ArrayExpress is a database of gene expression
and other microarray data at the European
Bioinformatics Institute (EBI)
www.ebi.ac.uk/arrayexpress/
Sample data set (from EBI ArrayExpress)
5
Obtaining data; an example
• Go to ArrayExpress and search “colon cancer.”
• Select Accession E-GEOD-42368, titled “p53dependent regulation of gene expression
following DNA damage” for Homo sapiens.
• Download the processed data as a zip file.
• Create a spreadsheet (e.g., Excel) and copy over
the data into it, one column per sample.
• Each column should have an ILMN_ID number,
and then for each sample, an expression level
and p-value.
• Organize the data by increasing p-values.
• Use david.abcc.ncifcrf.gov/ to locate gene
names from ILMN_IDs.
Preprocessing
Why Preprocessing?: The data may have nonbiological variation in the standardized data.





Thresholding
Scaling (log transformation)
Standardize
Normalization (Quantile Normalization)
Reducing the data set (by pairwise t-test)
7
Step 3: Formulate questions
about the data

Were there genes whose expression profiles
were correlated with colon cancer?

If so, how can we accurately determine which
of the samples are cancerous based entirely
on gene expression profiles?

Can any subtypes be identified by cluster
analysis across samples ?
8
Step 4: Computational and
statistical experiments
with R & Bioconductor

Class Prediction: Develop a multi-gene
predictor of class label for a sample using its
gene expression profile. (pairwise t-test)

Class Discovery: Use a various clustering
algorithms to discover clusters among samples
and genes. (K-means, hclust, PAM,…)
9
Hierarchical Clustering Results
Over expressed
in normal tissues
Over expressed
in cancer tissues
10
Gene 187 (Hsa.9972)
Step 5: Model for Cancer
Therapy
NCEH1
20X
ABCBs
2~3X
ABCB7
10X
Down-regulation of NCEH1 blocks cancer development?
11
Step 6: Validation of Model

Search PubMed for NCEH1 and cancer
http://www.ncbi.nlm.nih.gov/pubmed/17052608
12
Thank you!
13